v11.1.1

Merge pull request #933 from haverland/rolling
2025-12-07 12:06:58 +03:00 · 2022-08-22 18:20:00 +02:00 · 2022-08-22 18:12:08 +02:00 · 2022-08-22 17:58:30 +02:00 · 2022-08-22 16:10:07 +02:00 · 2022-08-22 10:33:10 +02:00
636 changed files with 52747 additions and 15100 deletions
--- a/.DS_Store
+++ b/.DS_Store
--- a/.gitignore
+++ b/.gitignore
@@ -4,6 +4,7 @@
 .code-workspace
 /sd-card/htm./.vscode/
 /code/build
+/sd-card/html/debug/

 CMakeLists.txt.user
 CMakeCache.txt
@@ -15,3 +16,5 @@ install_manifest.txt
 compile_commands.json
 CTestTestfile.cmake
 _deps
+code/edgeAI.code-workspace
+.DS_Store
--- a/Changelog.md
+++ b/Changelog.md
@@ -1,6 +1,245 @@
 # Versions

+##### 10.6.2 - Stability Increase (2022-07-24)

+- **NEW 10.6.2**: ignore hidden files in model selection (configuration page) 
+
+- **NEW 10.6.1**: Revoke esp32cam & tflite update
+
+- **NEW 10.6.1**: Bug Fix: tflite-filename with ".", HTML spelling error
+
+- IndluxDB: direct injection into InfluxDB - thanks to **[wetneb](https://github.com/wetneb)**
+
+- MQTT: implemented "Retain Flag" and extend with absolute Change (in addition to rate)
+
+- `config.ini`: removal of modelsize (readout from tflite)
+
+- Updated analog neural network file (`ana1000s2.tflite`) & digital neural network file (`dig1400s2q.tflite`)
+
+- TFMicro/Lite: Update (espressif Version 20220716)
+
+- Updated esp32cam (v20220716)
+
+- ESP-IDF: Update to 4.4
+
+- Internal update (CNN algorithm optimizations, reparation for new neural network type)
+
+- Bug Fix: no time with fixed IP, Postprocessing, MQTT
+
+  
+
+##### 10.5.2 - Stability Increase (2022-02-22)
+
+- NEW 10.5.2: Bug Fix: wrong `firmware.bin` (no rate update)
+- NEW 10.5.1: Bug Fix: wrong return value, rate value & PreValue status, HTML: SSID & IP were not displayed 
+- MQTT: changed wifi naming to "wifiRSSI"
+- HTML: check selectable values for consistency
+- Refactoring of check postprocessing consistency (e.g. max rate, negative rate, ...)
+- Bug Fix: corrected error in "Check Consistency Increase"
+
+
+
+##### 10.4.0 - Stability Increase (2022-02-12)
+
+- Graphical configuration: select available neural network files (*.tfl, *.tflite) from drop down menu
+- OTA-update: add option to upload tfl / tflite files to the correct location (`/config/`)
+  - In the future the new files will also be copied to the `firmware` directory of the repository
+- Added Wifi RSSI to MQTT information
+- Updated analog neural network file (`ana-s3-q-20220105.tflite`)
+- Updated digital neural network file (`dig-s1-q-20220102.tflite`)
+- Updated build environment to `Espressif 3.5.0`
+
+
+
+##### 10.3.0 - Stability Increase (2022-01-29)
+
+- Implemented LED flash dimming (`LEDIntensity`). 
+  Remark: as auto illumination in the camera is used, this is rather for energy saving. It will not help reducing reflections
+- Additional camera parameters: saturation, contrast (although not too much impact yet)
+- Some readings will have removable "N"s that can not be removed automatically and are handled with an "error" --> no return value in the field "value" anymore (still reported back via field "raw value")
+- Updated esp32 camera hardware driver
+- Bug fix: MQTT, HTML improvements
+
+**ATTENTION:  The new ESP32 camera hardware driver is much more stable on newer OV2640 versions (no or much less reboots) but seems to be not fully compatible with older versions.**
+
+* If you have problem with stalled systems you can try the following
+  - Update the parameter `ImageQuality` to `12` instead of current value `5` (manually in the `config.ini`)
+
+  - If this is not helping, you might need to update your hardware or stay with version 9.2
+
+##### 10.2.0 - Stability Increase (2022-01-14)
+
+- Due to the updated camera driver, the image looks different and a new setup might be needed
+
+  - Update reference image
+  - Update Alignment marks
+
+- Reduce reboot due to camera problems
+
+- Update esp32-camera to new version (master as of 2022-01-09)
+
+  
+
+##### 10.1.1 - Stability Increase (2022-01-12)
+
+- Bug Fix MQTT problem
+- Issue:
+  - Changing from v9.x to 10.x the MQTT-parameter "Topic" was renamed into "MainTopic" to address multiple number meters. This renaming should have been done automatically in the background within the graphical configuration, but was not working. Instead the parameter "Topic" was deleted and "MainTopic" was set to disabled and "undefined".
+- ToDo
+  - Update the `html.zip`
+  - If old `config.ini` available: copy it to `/config`, open the graphical configuration and save it again.
+  - If old `config.ini` not available: reset the parameter "MainTopic" within the `config.ini` manually
+  - Reboot
+
+##### 10.1.0 - Stability Increase (2022-01-09)
+
+- Reduce ESP32 frequency to 160MHz
+
+- Update tflite (new source: https://github.com/espressif/tflite-micro-esp-examples)
+
+- Update analog neural network (ana-s3-q-20220105.tflite)
+
+- Update digital neural network (dig-s1-q-20220102.tflite)
+
+- Increased web-server buffers
+- bug fix: compiler compatibility
+
+##### 10.0.2 - Stability Increase (2022-01-01)
+
+- NEW v10.0.2: Corrected JSON error
+
+- Updated compiler toolchain to ESP-IDF 4.3
+
+- Removal of memory leak
+
+- Improved error handling during startup (check PSRAM and camera with remark in logfile)
+
+- MQTT: implemented raw value additionally, removal of regex contrain
+
+- Normalized Parameter ``MaxRateValue``  to "change per minute" 
+
+- HTML: improved input handling
+
+- Corrected error handling: in case of error the old value, rate, timestamp are not transmitted any more
+
+  
+
+##### 9.2.0 - External Illumination (2021-12-02)
+
+- Direct JSON access: ``http://IP-ADRESS/json`` 
+- Error message in log file in case camera error during startup
+- Upgrade analog CNN to v9.1.0
+- Upgrade digital CNN to v13.3.0 (added new images)
+- html: support of different ports
+
+##### 9.1.1 - External Illumination (2021-11-16)
+
+- NEW 9.1.1 bug fix: LED implemenetation
+- External LEDs: change control mode (resolve bug with more than 2 LEDs)
+- Additional info into log file
+- Bug fix: decimal shift, html, log file
+
+##### 9.0.0 - External Illumination (2021-10-23)
+
+* Implementation of external illumination to adjust positioning, brightness and color of the illumination now set individually
+  * Technical details can be found in the wiki: https://github.com/jomjol/AI-on-the-edge-device/wiki/External-LED
+    <img src="https://raw.githubusercontent.com/jomjol/ai-on-the-edge-device/master/images/intern_vs_external.jpg" width="500">
+* New housing published for external LEDs and small clearing: https://www.thingiverse.com/thing:5028229
+
+
+
+##### 8.5.0 - Multi Meter Support (2021-10-07)
+
+* Upgrade digital CNN to v13.1.0 (added new images)
+* bug fix: wlan password with space, double digit output
+
+##### 8.4.0 - Multi Meter Support (2021-09-25)
+
+* License change (remove MIT license, remark see below)
+
+* html: show hostname in title and main page
+
+* configuration: 
+
+  * moved setting `ExtendedResolution` to individual number settings
+  * New parameter `IgnoreLeadingNaN` (delete leading NaN's specifically)
+  * **ATTENTION**: update of the `config.ini` needed (open, adjust `ExtendedResolution`, save)
+
+* Bug fixing (html, images of recognized numbers)
+
+  
+
+### **ATTENTION: LICENSE CHANGE - removal of MIT License.** 
+
+- Currently no licence published - copyright belongs to author
+- If you are interested in a commercial usage or dedicated versions please contact the developer
+  - no limits to private usage
+
+
+
+##### 8.3.0 - Multi Meter Support (2021-09-12)
+
+* Upgrade digital CNN to v12.1.0 (added new images)
+* Dedicated NaN handling, internal refactoring (CNN-Handling)
+* HTML: confirmation after config.ini update
+* Bug fixing
+
+##### 8.2.0 - Multi Meter Support (2021-08-24)
+
+* Improve server responsiveness
+* Flow status and prevalue status in overview
+* Improved prevalue handling 
+
+##### 8.1.0 - Multi Meter Support (2021-08-12)
+
+* GPIO: using the general mqtt main topic for GPIO
+
+* Upgrade digital CNN to v12.0.0  (added new images)
+* Update tfmicro to new master (2021-08-07)
+* Bug fix: remove text in mqtt value, remove connect limit in wlan reconnet
+
+##### 8.0.5 - Multi Meter Support (2021-08-01)
+
+* NEW 8.0.5: bug fix: saving prevalue
+* NEW 8.0.4: bug fix: load config.ini after upgrade
+* NEW 8.0.3: bug fix: reboot during `config.ini` handling, html error
+* NEW 8.0.2: saving roundes prevalue, bug fix html server
+* NEW 8.0.1: bug fix: html handling of parameter `FixedExposure` and `ImageSize`
+* Dual / multi meter support (more than 1 number to be recognized)
+  This is implemented with the feature "number" on the ROI definition as well as selected options
+* MQTT: standardization of the naming - including new topics (`json`,  `freeMem `, `uptime`)c
+* Preparation for extended GPIO support (thanks to Zwerk2k) - not tested and fully functional yet
+* Bug fixing: html server, memory leak, MQTT connect, hostname, turn of flash LED
+
+<span style="color: red;">**ATTENTION: the configuration and prevalue files are modified automatically and will not be backward compatible!**</span> 
+
+##### 7.1.2 MQTT-Update - (2021-06-17)
+
+* NEW: 7.1.2: bug fix setting hostname, Flash-LED not off during reboot
+
+* NEW: 7.1.1: bug fix wlan password with "="  (again)
+
+* MQTT error message: changes "no error", send retain flag
+
+* Update wlan handling to esp-idf 4.1
+
+* Upgrade digital CNN to v8.7.0  (added new images)
+
+* Bug fix: MQTT, WLAN, LED-Controll, GPIO usage, fixed IP, calculation flow rate
+
+  
+
+##### 7.0.1 MQTT-Update - (2021-05-13)
+
+* NEW: 7.0.1: bug fix wlan password with "=" 
+
+* Upgrade digital CNN to v8.5.0  (added new images)
+
+* New MQTT topics: flow rate (units/minute), time stamp (last correct read readout)
+
+* Update MQTT/Error topic to " " in case no error (instead of empty string)
+
+* Portrait or landscape image orientation in rotated image (avoid cropping)

 ##### 6.7.2 Image Processing in Memory - (2021-05-01)

--- a/FeatureRequest.md
+++ b/FeatureRequest.md
@@ -11,6 +11,137 @@

 ____

+#### #29 Add favicon and use the hostname for the website
+
+* https://github.com/jomjol/AI-on-the-edge-device/issues/927
+
+#### #28 Improved error handling for ROIs
+
+* In case a ROI is out of the image, there is no error message, but a non sense image is used
+* Implement a error message for wrong configuratioin of ROI
+
+#### #27 Use Homie Spec for Mqtt binding
+
+* Use the standardized Home Protocol for the Mqtt binding 
+* https://homieiot.github.io/
+
+#### #26 Changes behaviour for "N" replacement
+
+* in case the higher digits has already increased by minium 1 - don't set the "N" to the last value, but to "0"
+* https://github.com/jomjol/AI-on-the-edge-device/issues/792
+
+
+#### #25 Trigger Measurement via MQTT
+
+* https://github.com/jomjol/AI-on-the-edge-device/issues/727
+
+
+#### #24 Show Mqtt state directly in Webserver
+
+* Show MQTT log in Web page. E.g. connection established or failed to connect...
+
+  
+
+
+#### #23 CPU Temp and Mqtt values
+
+* Show the CPU Temp directly in Webpage. Also add the value to MQTT sending
+
+  
+
+#### #22 Direct hint to the different neural network files in the other repositories
+
+* https://github.com/jomjol/AI-on-the-edge-device/issues/644
+
+  
+
+#### #21 Extended "CheckDigitalConsistency" Logik
+
+* https://github.com/jomjol/AI-on-the-edge-device/issues/590
+
+  
+
+#### #20 Deep sleep and push mode
+
+* Let the device be normally in deep sleep state, and wake it up periodically to collect data and push it via MQTT or HTTP post.
+* Support ESP-NOW to reduce the overhead of connecting to wifi and mqtt 
+* the above should enable battery powered applications
+  
+
+#### #19 Extended log informations
+
+* https://github.com/jomjol/AI-on-the-edge-device/issues/580
+
+  
+
+#### ~~#18 Document WLAN-strength in web page~~
+
+* ~~https://github.com/jomjol/AI-on-the-edge-device/issues/563~~
+
+
+
+#### ~~#17 Direct InfluxDB connection~~
+
+* ~~Done in v10.6.0~~
+
+
+#### #16 Serial Communication
+
+* https://github.com/jomjol/AI-on-the-edge-device/issues/512
+* Send the readout value via RX/TX interface with a dedicated TAG
+* Make dedicated communication FlowModule
+* Modification of RX/TX communication
+* Configuration interfache
+
+
+#### #15 Calibration for FishEye image
+
+* https://github.com/jomjol/AI-on-the-edge-device/issues/507
+
+1.  The development of such a correction algorithm with the libraries, that are available for the ESP32 environment.
+2. New module for integration of the flow into the image processing flow.
+3. Extension of the configuration (config.ini) and html-pages
+4. Parameter adjustment and testing for every different fish-eye module
+5. Maintenance for further updates / modules, ...
+
+
+
+#### #14 Backup and restore option for configuration
+
+* https://github.com/jomjol/AI-on-the-edge-device/issues/459
+
+* Implement a zip file compression for store and restore
+
+* Update the html to handle it
+
+  
+
+#### #13 Manage non linear gauge without CNN re-training
+
+* https://github.com/jomjol/AI-on-the-edge-device/issues/443
+
+* Implement a look up table for non linear analog meters
+
+  
+
+#### ~~#12 Less reboots due to memory leakage~~
+
+* ~~Issue: #414 & #425  #430~~
+
+  
+
+#### #11 MQTT - configurable payload
+
+* https://github.com/jomjol/AI-on-the-edge-device/issues/344
+
+  
+
+#### #10 Improve and bug fix logging of images
+
+* https://github.com/jomjol/AI-on-the-edge-device/issues/307
+
+  
+
 #### #9 Basic auth for the UI

 * https://github.com/jomjol/AI-on-the-edge-device/issues/283
@@ -114,4 +245,4 @@ ____

 * ~~Implementation of a software module for external light source (e.g. WS8132 LED controller, ...)~~
 * ~~Update of the camera module to use the external light instead of the internal flash light~~
-* ~~Adopt the configuration algorithm with a configurable light source~~
+* ~~Adopt the configuration algorithm with a configurable light source~~
--- a/21
+++ b/21
@@ -1,21 +0,0 @@
-MIT License
-
-Copyright (c) 2020 jomjol
-
-Permission is hereby granted, free of charge, to any person obtaining a copy
-of this software and associated documentation files (the "Software"), to deal
-in the Software without restriction, including without limitation the rights
-to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
-copies of the Software, and to permit persons to whom the Software is
-furnished to do so, subject to the following conditions:
-
-The above copyright notice and this permission notice shall be included in all
-copies or substantial portions of the Software.
-
-THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
-IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
-FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
-AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
-LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
-OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
-SOFTWARE.
--- a/README.md
+++ b/README.md
@@ -4,9 +4,10 @@ This is an example of Artificial Intelligence (AI) calculations on a very cheap

 ### Details on **function**, **installation** and **configuration** can be found on the **[Wiki Page](https://github.com/jomjol/AI-on-the-edge-device/wiki)**

-A 3d-printable housing can be found here: https://www.thingiverse.com/thing:4573481
-
-respectively ESP32-Cam housing only: https://www.thingiverse.com/thing:4571627
+A 3d-printable housing can be found here:
+  - https://www.thingiverse.com/thing:4573481 (Water Meter)
+  - https://www.thingiverse.com/thing:5028229 (Power Meter)
+  - https://www.thingiverse.com/thing:4571627 (ESP32-Cam housing only)

 <img src="https://raw.githubusercontent.com/jomjol/AI-on-the-edge-device/master/images/watermeter_all.jpg" width="200"><img src="https://raw.githubusercontent.com/jomjol/AI-on-the-edge-device/master/images/main.jpg" width="200"><img src="https://raw.githubusercontent.com/jomjol/AI-on-the-edge-device/master/images/size.png" width="200"> 

@@ -32,66 +33,48 @@ If you have any technical topics, you can file a issue in this repository.

 In other cases you can contact the developer via email: <img src="https://raw.githubusercontent.com/jomjol/AI-on-the-edge-device/master/images/mail.jpg" height="25"> 

+
+------
 ## Change log
+**General remark:** Besides the file `firmware.bin`, typically the content of `/html` will need to be updated!

 ------

-### Known Issues
+##### 11.1.1 - Intermediate Digits (2022-08-22)

-* slow response of web server during picture analysis
-* spontaneous reboots (mostly due to html access during image processing) - self recovery implemented
+- New and improved consistency check (especially with analog and digital counters mixed)
+- Bug Fix: digital counter algorithm

------
+##### 11.0.1 - Intermediate Digits (2022-08-18)

-**General remark:** Beside the `firmware.bin`, typically also the content of `/html` needs to be updated!
+- **NEW v11.0.1**: Bug Fix InfluxDB configuration (only update of html.zip necessary)

+- Implementation of new CNN types to detect intermediate values of digits with rolling numbers

+  - By default the old algo (0, 1, ..., 9, "N") is active (due to the limited types of digits trained so far)
+  - Activation can be done by selection a tflite file with the new trained model in the 'config.ini'
+  - **Details can be found in the [wiki](https://github.com/jomjol/AI-on-the-edge-device/wiki/Neural-Network-Types)** (different types, trained image types, naming convention)

-##### 8.0.5 - Multi Meter Support (2021-07-29)
+- Updated  neural network files (and adaption to new naming convention)

-* NEW 8.0.5: bug fix: saving prevalue
-* NEW 8.0.4: bug fix: load config.ini after upgrade
-* NEW 8.0.3: bug fix: reboot during `config.ini` handling, html error
-* NEW 8.0.2: saving roundes prevalue, bug fix html server
-* NEW 8.0.1: bug fix: html handling of parameter `FixedExposure` and `ImageSize`
-* Dual / multi meter support (more than 1 number to be recognized)
-  This is implemented with the feature "number" on the ROI definition as well as selected options
-* MQTT: standardization of the naming - including new topics (`json`,  `freeMem `, `uptime`)c
-* Preparation for extended GPIO support (thanks to Zwerk2k) - not tested and fully functional yet
-* Bug fixing: html server, memory leak, MQTT connect, hostname, turn of flash LED
+- Published a tool to download and combine log files - **Thanks to **

-<span style="color: red;">**ATTENTION: the configuration and prevalue files are modified automatically and will not be backward compatible!**</span> 
+  - Files see ['/tools/logfile-tool'](tbd), How-to see [wiki](https://github.com/jomjol/AI-on-the-edge-device/wiki/Gasmeter-Log-Downloader)

-
-
-##### 7.1.2 MQTT-Update - (2021-06-17)
-
-* NEW: 7.1.2: bug fix setting hostname, Flash-LED not off during rebootNEW: 7.1.1: bug fix wlan password with "="  (again)
-* MQTT error message: changes "no error", send retain flag
-* Update wlan handling to esp-idf 4.1
-* Upgrade digital CNN to v8.7.0  (added new images)
-* Bug fix: MQTT, WLAN, LED-Controll, GPIO usage, fixed IP, calculation flow rate
+- Bug Fix: InfluxDB enabling in grahic configuration

  

-##### 7.0.1 MQTT-Update - (2021-05-13)
+## Tools

-* NEW: 7.0.1: bug fix wlan password with "=" 
-
-* Upgrade digital CNN to v8.5.0  (added new images)
-
-* New MQTT topics: flow rate (units/minute), time stamp (last correct read readout)
-
-* Update MQTT/Error topic to " " in case no error (instead of empty string)
-
-* Portrait or landscape image orientation in rotated image (avoid cropping)
-
-  
+* Logfile downloader and combiner (Thx to [reserve85](https://github.com/reserve85))
+  * Files see ['/tools/logfile-tool'](tbd), How-to see [wiki](https://github.com/jomjol/AI-on-the-edge-device/wiki/Gasmeter-Log-Downloader)


-## Additional ideas

-There are some ideas and feature request, which are not followed currently - mainly due to capacity reasons on side of the developer. They are collected here: [FeatureRequest.md](FeatureRequest.md)
+## Additional Ideas
+
+There are some ideas and feature requests which are not followed currently - mainly due to capacity reasons on side of the developer. They are collected here: [FeatureRequest.md](FeatureRequest.md)



@@ -99,6 +82,14 @@ There are some ideas and feature request, which are not followed currently - mai

 ## History

+##### 10.6.2 - Stability Increase (2022-07-24)
+
+##### 9.2.0 - External Illumination (2021-12-02)
+
+##### 8.5.0 Multi Meter Support (2021-10-07)
+
+##### 7.1.2 MQTT-Update - (2021-06-17)
+
 ##### 6.7.2 Image Processing in Memory - (2021-05-01)

 ##### 5.0.0 Setup Modus - (2020-12-06)
@@ -121,8 +112,3 @@ There are some ideas and feature request, which are not followed currently - mai

 #### [Full Changelog](Changelog.md)

-
-
-## Solved topics
-
-* n.a.
--- a/code/.helper/copy-esp-idf.bat
+++ b/code/.helper/copy-esp-idf.bat
@@ -0,0 +1,3 @@
+copy "..\..\code\build\esp32cam-server-only.bin" "..\..\firmware\firmware.bin"
+copy "..\..\code\build\bootloader\bootloader.bin" "..\..\firmware\bootloader.bin"
+copy "..\..\code\build\partition_table\partition-table.bin" "..\..\firmware\partitions.bin"
--- a/code/components/esp-nn/.gitignore
+++ b/code/components/esp-nn/.gitignore
@@ -0,0 +1,57 @@
+.config
+*.o
+*.i
+*.s
+*.orig
+*.pyc
+
+# gtags
+GTAGS
+GRTAGS
+GPATH
+
+# emacs
+.dir-locals.el
+
+# emacs temp file suffixes
+*~
+.#*
+\#*#
+
+# eclipse setting
+.settings
+
+# MacOS directory files
+.DS_Store
+
+# Example project files
+examples/**/sdkconfig
+examples/**/sdkconfig.old
+examples/**/build
+
+# Test app files
+test_app/build
+test_app/sdkconfig
+test_app/sdkconfig.old
+
+# Doc build artifacts
+docs/_build/
+docs/doxygen-warning-log.txt
+docs/sphinx-warning-log.txt
+docs/sphinx-warning-log-sanitized.txt
+docs/xml/
+docs/xml_in/
+docs/man/
+docs/doxygen_sqlite3.db
+
+TEST_LOGS
+
+
+# gcov coverage reports
+*.gcda
+*.gcno
+coverage.info
+coverage_report/
+
+# VS Code Settings
+.vscode/
--- a/code/components/esp-nn/.gitlab-ci.yml
+++ b/code/components/esp-nn/.gitlab-ci.yml
@@ -0,0 +1,55 @@
+stages:
+  - build
+
+variables:
+  BATCH_BUILD: "1"
+  V: "0"
+  MAKEFLAGS: "-j8 --no-keep-going"
+  IDF_PATH: "$CI_PROJECT_DIR/esp-idf"
+  LOG_PATH: "$CI_PROJECT_DIR"
+
+.set_git_config: &set_git_config
+  # Set git config
+  - git config user.email "test@espressif.com"
+  - git config user.name "Espressif"
+
+.add_ssh_key: &add_ssh_key
+  # Add gitlab ssh key
+  - mkdir -p ~/.ssh
+  - chmod 700 ~/.ssh
+  - echo -n $GITLAB_KEY > ~/.ssh/id_rsa_base64
+  - base64 --decode --ignore-garbage ~/.ssh/id_rsa_base64 > ~/.ssh/id_rsa
+  - chmod 600 ~/.ssh/id_rsa
+  - echo -e "Host gitlab.espressif.cn\n\tStrictHostKeyChecking no\n" >> ~/.ssh/config
+
+before_script:
+  # Add gitlab ssh key
+  - *add_ssh_key
+  # Set git config
+  - *set_git_config
+
+.build_esp32s3: &build_esp32s3
+  - idf.py set-target esp32s3 build
+
+.build_esp32: &build_esp32
+  - idf.py set-target esp32 build
+
+build_demo:
+  stage: build
+  image: $CI_DOCKER_REGISTRY/esp32-ci-env:esp-nn
+  tags:
+    - build
+  script:
+    # Clone IDF
+    - git clone --recursive --single-branch -b release/v4.4 --reference-if-able /local_references/gitlab/ https://gitlab-ci-token:${BOT_TOKEN}@gitlab.espressif.cn:6688/espressif/esp-idf.git
+    - cd esp-idf
+    - ./install.sh
+    - . ./export.sh
+    - cd ..
+    # Build examples now
+    - cd test_app
+    # Build esp32s3
+    - *build_esp32s3
+    # Build esp32
+    - *build_esp32
+    - cd -
--- a/code/components/esp-nn/CMakeLists.txt
+++ b/code/components/esp-nn/CMakeLists.txt
@@ -0,0 +1,48 @@
+idf_build_get_property(idf_target IDF_TARGET)
+
+set(c_srcs
+    "src/activation_functions/esp_nn_relu_ansi.c"
+    "src/basic_math/esp_nn_add_ansi.c"
+    "src/basic_math/esp_nn_mul_ansi.c"
+    "src/convolution/esp_nn_conv_ansi.c"
+    "src/convolution/esp_nn_depthwise_conv_ansi.c"
+    "src/fully_connected/esp_nn_fully_connected_ansi.c"
+    "src/softmax/esp_nn_softmax_ansi.c"
+    "src/softmax/esp_nn_softmax_opt.c"
+    "src/pooling/esp_nn_avg_pool_ansi.c"
+    "src/pooling/esp_nn_max_pool_ansi.c")
+
+if(CONFIG_IDF_TARGET_ESP32S3)
+    set(s3_srcs
+        "src/common/esp_nn_common_functions_esp32s3.S"
+        "src/common/esp_nn_multiply_by_quantized_mult_esp32s3.S"
+        "src/common/esp_nn_multiply_by_quantized_mult_ver1_esp32s3.S"
+        "src/activation_functions/esp_nn_relu_s8_esp32s3.S"
+        "src/basic_math/esp_nn_add_s8_esp32s3.S"
+        "src/basic_math/esp_nn_mul_s8_esp32s3.S"
+        "src/convolution/esp_nn_conv_esp32s3.c"
+        "src/convolution/esp_nn_depthwise_conv_s8_esp32s3.c"
+        "src/convolution/esp_nn_conv_s16_mult8_esp32s3.S"
+        "src/convolution/esp_nn_conv_s16_mult8_1x1_esp32s3.S"
+        "src/convolution/esp_nn_conv_s16_mult4_1x1_esp32s3.S"
+        "src/convolution/esp_nn_depthwise_conv_s8_mult1_3x3_padded_esp32s3.S"
+        "src/convolution/esp_nn_depthwise_conv_s16_mult1_esp32s3.S"
+        "src/convolution/esp_nn_depthwise_conv_s16_mult1_3x3_esp32s3.S"
+        "src/convolution/esp_nn_depthwise_conv_s16_mult1_3x3_no_pad_esp32s3.S"
+        "src/convolution/esp_nn_depthwise_conv_s16_mult8_3x3_esp32s3.S"
+        "src/convolution/esp_nn_depthwise_conv_s16_mult4_esp32s3.S"
+        "src/convolution/esp_nn_depthwise_conv_s16_mult8_esp32s3.S"
+        "src/fully_connected/esp_nn_fully_connected_s8_esp32s3.S"
+        "src/pooling/esp_nn_max_pool_s8_esp32s3.S"
+        "src/pooling/esp_nn_avg_pool_s8_esp32s3.S")
+endif()
+
+idf_component_register(SRCS "${c_srcs}"
+                            "${s3_srcs}"
+                       INCLUDE_DIRS "include" "src/common")
+
+if(CONFIG_IDF_TARGET_ESP32S3)
+    target_compile_options(${COMPONENT_LIB} PRIVATE -mlongcalls -fno-unroll-loops -O2 -Wno-unused-function)
+else()
+    target_compile_options(${COMPONENT_LIB} PRIVATE -Wno-unused-function)
+endif()
--- a/code/components/esp-nn/Kconfig.projbuild
+++ b/code/components/esp-nn/Kconfig.projbuild
@@ -0,0 +1,29 @@
+menu "ESP-NN"
+
+choice NN_OPTIMIZATIONS
+   bool "Optimization for nn functions"
+   default NN_OPTIMIZED
+   help
+      Use ANSI-C versions for verification and debug purpose.
+      Optimisations are automatically picked up for a chipset.
+      For ESP32-S3, assembly Optimisations are selected.
+      For ESP32, just the ANSI C versions are selected for now.
+
+config NN_ANSI_C
+   bool "ANSI C"
+   help
+      ANSI C versions for verification and debug purposes.
+config NN_OPTIMIZED
+   bool "Optimized versions"
+   help
+      Optimisations are automatically picked up for a chipset.
+      For ESP32-S3, assembly Optimisations are selected.
+      For ESP32, just the ANSI C versions are selected for now.
+endchoice
+
+config NN_OPTIMIZATIONS
+   int
+   default 0 if NN_ANSI_C
+   default 1 if NN_OPTIMIZED
+
+endmenu
--- a/code/components/tfmicro/third_party/gemmlowp/LICENSE
+++ b/code/components/tfmicro/third_party/gemmlowp/LICENSE
--- a/code/components/esp-nn/README.md
+++ b/code/components/esp-nn/README.md
@@ -0,0 +1,54 @@
+# ESP-NN
+
+The library contains optimised NN (Neural Network) functions for various Espressif chipsets.
+
+* Supported platforms:
+   * TensorFlow Lite Micro (TFLite Micro). Repo can be found [here](https://github.com/espressif/tflite-micro-esp-examples)
+
+* Supported ESP chipsets include:
+   * ESP32-S3 (Assembly versions optimised to benefit from vector instructions of ESP32-S3)
+   * ESP32 (ANSI C versions)
+
+## Performance
+
+### Kernelwise performance for s8 versions:
+
+  * Kernelwise performance on ESP32-S3 chip
+    * Numbers are ticks taken for kernel to execute
+    * Chip config: 240MHz, SPI: QPI 80MHz, Data cache: 64KB
+
+    | Function        | ANSI C  | ESP32-S3 Opt  | Opt Ratio | Data info   | Memory    |
+    | ----------------| --------|---------|---------|-------------|-----------|
+    | elementwise_add | 320397  | 87119   | 3.68    | size = 1615 | External  |
+    | elementwise_mul | 125958  | 44239   | 2.85    | size = 1615 | External  |
+    | convolution     | 4663012 | 428675  | 10.88   | input(10,10), filter(64x1x1x64) | External |
+    | convolution     | 301014  | 32433   | 9.28    | input(8,8), filter(16x1x1x16) | External |
+    | convolution     | 2115418 | 1020923 | 2.07    | input(10,10), filter(64x3x3x3) | External |
+    | depthwise conv  | 1190062 | 203278  | 5.85    | input (18, 18), pad(0,0), stride(1,1) filter: 1x3x3x16 | External |
+    | depthwise conv  | 837072  | 182335  | 4.59    | input (12, 12), pad(1,1), stride(1,1)  filter: 8x5x5x4 | External |
+    | max pool        | 485714  | 76747   | 6.33    | input(16,16), filter (1x3x3x16) | Internal |
+    | avg pool        | 541462  | 160580  | 3.37    | input(16,16), filter (1x3x3x16) | Internal |
+    | fully connected | 15853   | 9547    | 1.66    | len: 265, ch = 3 | Internal |
+    | prelu (relu6)   | 19472   | 2734    | 7.12    | size, 1615  | Internal  |
+
+
+## Configuration
+
+  * To configure, please use `idf.py menuconfig` and under `ESP-NN` select `NN_OPTIMIZATIONS`
+  * There are two options presented:
+     * Optimized versions
+     * ANSI C
+
+  * Default selection is for `Optimized versions`. For ESP32-S3, assembly versions are automatically selected, whereas for ESP32,  ANSI-C versions are selected by default.
+  * For debugging purposes, you may want to select `ANSI C`
+
+
+## Contributing
+
+If you encounter an issue with ESP-NN, or wish to submit a feature request, please use the Issues section on the Github.
+
+For general questions related to this library, please use the esp32.com forum.
+
+## Copyrights and License
+
+All original source code in this repository is Copyright (C) 2020-2021 Espressif Systems. This source code is licensed under the Apache License 2.0 as described in the file LICENSE.
--- a/code/components/esp-nn/include/esp_nn.h
+++ b/code/components/esp-nn/include/esp_nn.h
@@ -0,0 +1,46 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#if defined(CONFIG_NN_OPTIMIZED)
+#ifdef CONFIG_IDF_TARGET_ESP32S3
+#define ARCH_ESP32_S3 1
+#endif
+#ifdef CONFIG_IDF_TARGET_ESP32
+#define ARCH_ESP32 1
+#endif
+#endif
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/* reference kernels included by default */
+#include "esp_nn_ansi_headers.h"
+
+#if defined(CONFIG_NN_OPTIMIZED)
+#ifdef ARCH_ESP32_S3
+#include "esp_nn_esp32s3.h"
+#endif
+#ifdef ARCH_ESP32
+#include "esp_nn_esp32.h"
+#endif
+#else
+#include "esp_nn_ansi_c.h"
+#endif
+
+#ifdef __cplusplus
+}
+#endif
--- a/code/components/esp-nn/include/esp_nn_ansi_c.h
+++ b/code/components/esp-nn/include/esp_nn_ansi_c.h
@@ -0,0 +1,46 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+/**
+ * @file        Header definitions to include for ANSI C versions.
+ *              These are just typedefs to pick up ANSI versions.
+ */
+
+#pragma once
+
+#include "esp_nn_ansi_headers.h"
+
+#define esp_nn_add_elementwise_s8 esp_nn_add_elementwise_s8_ansi
+#define esp_nn_mul_elementwise_s8 esp_nn_mul_elementwise_s8_ansi
+
+#define esp_nn_depthwise_conv_s8 esp_nn_depthwise_conv_s8_ansi
+
+#define esp_nn_conv_s8 esp_nn_conv_s8_ansi
+
+#define esp_nn_get_conv_scratch_size esp_nn_get_conv_scratch_size_ansi
+#define esp_nn_set_conv_scratch_buf esp_nn_set_conv_scratch_buf_ansi
+
+#define esp_nn_get_depthwise_conv_scratch_size esp_nn_get_depthwise_conv_scratch_size_ansi
+#define esp_nn_set_depthwise_conv_scratch_buf esp_nn_set_depthwise_conv_scratch_buf_ansi
+
+#define esp_nn_relu6_s8 esp_nn_relu6_s8_ansi
+
+#define esp_nn_avg_pool_s8 esp_nn_avg_pool_s8_ansi
+#define esp_nn_max_pool_s8 esp_nn_max_pool_s8_ansi
+
+#define esp_nn_fully_connected_s8 esp_nn_fully_connected_s8_ansi
+
+#define esp_nn_get_softmax_scratch_size esp_nn_get_softmax_scratch_size_ansi
+#define esp_nn_set_softmax_scratch_buf esp_nn_set_softmax_scratch_buf_ansi
+#define esp_nn_softmax_s8 esp_nn_softmax_s8_ansi
--- a/code/components/esp-nn/include/esp_nn_ansi_headers.h
+++ b/code/components/esp-nn/include/esp_nn_ansi_headers.h
@@ -0,0 +1,283 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+/**
+ * @file        Header definitions to include for esp_nn reference functions
+ */
+
+#include <stdint.h>
+
+/************************** Basic math functions ****************************/
+
+/**
+ * @brief       elementwise addition
+ *
+ * @note        inputs type: int8_t, output: int8_t
+ *              input offsets: although int32_t, they are contained in 8 bits [-128, 127]
+ *
+ *              shift values are expected to be <= 0
+ */
+void esp_nn_add_elementwise_s8_ansi(const int8_t *input1_data,
+                                    const int8_t *input2_data,
+                                    const int32_t input1_offset,
+                                    const int32_t input2_offset,
+                                    const int32_t input1_mult,
+                                    const int32_t input2_mult,
+                                    const int32_t input1_shift,
+                                    const int32_t input2_shift,
+                                    const int32_t left_shift,
+                                    int8_t *output,
+                                    const int32_t out_offset,
+                                    const int32_t out_mult,
+                                    const int32_t out_shift,
+                                    const int32_t activation_min,
+                                    const int32_t activation_max,
+                                    const int32_t size);
+/**
+ * @brief       elementwise multiplication
+ *
+ * @note        inputs type: int8_t, output: int8_t
+ *              input offsets: although int32_t, they are contained in 8 bits [-128, 127]
+ *
+ *              output shift is expected to be <= 0
+ */
+void esp_nn_mul_elementwise_s8_ansi(const int8_t *input1_data,
+                                    const int8_t *input2_data,
+                                    const int32_t input1_offset,
+                                    const int32_t input2_offset,
+                                    int8_t *output,
+                                    const int32_t out_offset,
+                                    const int32_t out_mult,
+                                    const int32_t out_shift,
+                                    const int32_t activation_min,
+                                    const int32_t activation_max,
+                                    const int32_t size);
+
+
+/************************** Convolution functions *****************************/
+
+/**
+ * @brief       depthwise convolution per channel
+ *
+ * @note        inputs type: int8_t, output: int8_t
+ *              Version used in tflite is per channel.
+ *              This version follows the same footsprints.
+ *              Meaning, it has per out_channel shift and multiplier for
+ *              requantization
+ *
+ *              optimization notes: Though input_offset is int32 type,
+ *              offset values are contained in 8 bits [-128, 127]
+ */
+void esp_nn_depthwise_conv_s8_ansi(const int8_t *input_data,
+                                   const uint16_t input_wd,
+                                   const uint16_t input_ht,
+                                   const uint16_t channels,
+                                   const int32_t input_offset,
+                                   const uint16_t pad_wd,
+                                   const uint16_t pad_ht,
+                                   const uint16_t stride_wd,
+                                   const uint16_t stride_ht,
+                                   const uint16_t ch_mult,
+                                   const int8_t *filter_data,
+                                   const uint16_t filter_wd,
+                                   const uint16_t filter_ht,
+                                   const int32_t *bias,
+                                   int8_t *out_data,
+                                   const uint16_t out_wd,
+                                   const uint16_t out_ht,
+                                   const int32_t out_offset,
+                                   const int32_t *out_shift,
+                                   const int32_t *out_mult,
+                                   const int32_t activation_min,
+                                   const int32_t activation_max);
+
+/**
+ * @brief       2d-convolution channelwise
+ *
+ * @note        operation: result += (input + offset) * filter
+ *
+ *              inputs type: int8_t, output: int8_t
+ *              input offsets: although int32_t, they are contained in 8 bits [-128, 127]
+ */
+void esp_nn_conv_s8_ansi(const int8_t *input_data,
+                         const uint16_t input_wd,
+                         const uint16_t input_ht,
+                         const uint16_t in_channels,
+                         const int32_t input_offset,
+                         const uint16_t pad_wd,
+                         const uint16_t pad_ht,
+                         const uint16_t stride_wd,
+                         const uint16_t stride_ht,
+                         const int8_t *filter_data,
+                         const uint16_t filter_wd,
+                         const uint16_t filter_ht,
+                         const int32_t *bias,
+                         int8_t *out_data,
+                         const uint16_t out_wd,
+                         const uint16_t out_ht,
+                         const uint16_t out_channels,
+                         const int32_t out_offset,
+                         const int32_t *out_shift,
+                         const int32_t *out_mult,
+                         const int32_t activation_min,
+                         const int32_t activation_max);
+
+int esp_nn_get_conv_scratch_size_ansi(const uint16_t input_wd,
+                                      const uint16_t input_ht,
+                                      const uint16_t in_ch,
+                                      const uint16_t out_ch,
+                                      const uint16_t filter_wd,
+                                      const uint16_t filter_ht);
+void esp_nn_set_conv_scratch_buf_ansi(const void *buf);
+
+int esp_nn_get_depthwise_conv_scratch_size_ansi(const uint16_t input_wd,
+                                                const uint16_t input_ht,
+                                                const uint16_t channels,
+                                                const uint16_t ch_mult,
+                                                const uint16_t filter_wd,
+                                                const uint16_t filter_ht);
+void esp_nn_set_depthwise_conv_scratch_buf_ansi(const void *buf);
+
+/************************** Activation functions *****************************/
+
+/**
+ * @brief       relu6
+ *
+ * @note        inout: int8_t
+ */
+void esp_nn_relu6_s8_ansi(int8_t *data, uint16_t size);
+
+/************************** Pooling functions *****************************/
+
+
+/**
+ * @brief       max_pool
+ *
+ * @note        inputs type: int8_t, output: int8_t
+ *              input offsets: although int32_t, they are contained in 8 bits [-128, 127]
+ */
+void esp_nn_max_pool_s8_ansi(const int8_t *input,
+                             const uint16_t input_wd,
+                             const uint16_t input_ht,
+                             int8_t *output,
+                             const uint16_t output_wd,
+                             const uint16_t output_ht,
+                             const uint16_t stride_wd,
+                             const uint16_t stride_ht,
+                             const uint16_t filter_wd,
+                             const uint16_t filter_ht,
+                             const uint16_t pad_wd,
+                             const uint16_t pad_ht,
+                             const int32_t activation_min,
+                             const int32_t activation_max,
+                             const uint16_t channels);
+
+/**
+ * @brief       avg_pool
+ *
+ * @note        inputs type: int8_t, output: int8_t
+ *              input offsets: although int32_t, they are contained in 8 bits [-128, 127]
+ */
+void esp_nn_avg_pool_s8_ansi(const int8_t *input,
+                             const uint16_t input_wd,
+                             const uint16_t input_ht,
+                             int8_t *output,
+                             const uint16_t output_wd,
+                             const uint16_t output_ht,
+                             const uint16_t stride_wd,
+                             const uint16_t stride_ht,
+                             const uint16_t filter_wd,
+                             const uint16_t filter_ht,
+                             const uint16_t pad_wd,
+                             const uint16_t pad_ht,
+                             const int32_t activation_min,
+                             const int32_t activation_max,
+                             const uint16_t channels);
+
+
+/************************** Fully connected functions ***********************/
+
+/**
+ * @brief       fully connected
+ *
+ * @note        inputs type: int8_t, output: int8_t
+ *              input offsets: although int32_t, they are contained in 8 bits [-128, 127]
+ */
+void esp_nn_fully_connected_s8_ansi(const int8_t *input_data,
+                                    const int32_t input_offset,
+                                    const uint16_t row_len,
+                                    const int8_t *filter_data,
+                                    const int32_t filter_offset,
+                                    const int32_t *bias,
+                                    int8_t *out_data,
+                                    const uint16_t out_channels,
+                                    const int32_t out_offset,
+                                    const int32_t out_shift,
+                                    const int32_t out_mult,
+                                    const int32_t activation_min,
+                                    const int32_t activation_max);
+
+/**
+ * @brief   Get scratch buffer size needed by softmax function
+ *
+ * @param   width
+ * @param   height
+ * @return  size in bytes
+ *
+ * @note    buffer must be 4 byte aligned
+ */
+int32_t esp_nn_get_softmax_scratch_size_ansi(const int32_t width, const int32_t height);
+
+/* ANSI C function to be hooked up when optimised version needed */
+int32_t esp_nn_get_softmax_scratch_size_opt(const int32_t width, const int32_t height);
+
+/**
+ * @brief   Set scratch buffer to be used by softmax function
+ *
+ * @param   buffer  this can be NULL if one needs to unset it
+ *                  must be aligned to 4 bytes
+ */
+void esp_nn_set_softmax_scratch_buf_ansi(void *buffer);
+
+/* ANSI C function to be hooked up when optimised version needed */
+void esp_nn_set_softmax_scratch_buf_opt(void *buffer);
+
+/**
+ * @brief       reference softmax function
+ *
+ * @note        inputs type: int8_t, output: int8_t
+ */
+void esp_nn_softmax_s8_ansi(const int8_t *input_data,
+                            const int32_t height,
+                            const int32_t width,
+                            const int32_t mult,
+                            const int32_t shift,
+                            const int32_t diff_min,
+                            int8_t *output_data);
+
+/**
+ * @brief       optimised version of softmax function
+ *
+ * @note        the function uses extra buffer (4 * width bytes)
+ *              hence, scratch buffers must be set before calling this.
+ */
+void esp_nn_softmax_s8_opt(const int8_t *input_data,
+                           const int32_t height,
+                           const int32_t width,
+                           const int32_t mult,
+                           const int32_t shift,
+                           const int32_t diff_min,
+                           int8_t *output_data);
--- a/code/components/esp-nn/include/esp_nn_esp32.h
+++ b/code/components/esp-nn/include/esp_nn_esp32.h
@@ -0,0 +1,48 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+/**
+ * @file        Header definitions to include for esp_nn optimized functions for
+ *              the ESP32 platform.
+ *              We are hooking up just the C versions for now.
+ *              The file hence is exactly same as `esp_nn_ansi_c.h`
+ */
+
+#pragma once
+
+#include "esp_nn_ansi_headers.h"
+
+#define esp_nn_add_elementwise_s8 esp_nn_add_elementwise_s8_ansi
+#define esp_nn_mul_elementwise_s8 esp_nn_mul_elementwise_s8_ansi
+
+#define esp_nn_depthwise_conv_s8 esp_nn_depthwise_conv_s8_ansi
+
+#define esp_nn_conv_s8 esp_nn_conv_s8_ansi
+
+#define esp_nn_get_conv_scratch_size esp_nn_get_conv_scratch_size_ansi
+#define esp_nn_set_conv_scratch_buf esp_nn_set_conv_scratch_buf_ansi
+
+#define esp_nn_get_depthwise_conv_scratch_size esp_nn_get_depthwise_conv_scratch_size_ansi
+#define esp_nn_set_depthwise_conv_scratch_buf esp_nn_set_depthwise_conv_scratch_buf_ansi
+
+#define esp_nn_relu6_s8 esp_nn_relu6_s8_ansi
+
+#define esp_nn_avg_pool_s8 esp_nn_avg_pool_s8_ansi
+#define esp_nn_max_pool_s8 esp_nn_max_pool_s8_ansi
+
+#define esp_nn_fully_connected_s8 esp_nn_fully_connected_s8_ansi
+
+#define esp_nn_get_softmax_scratch_size esp_nn_get_softmax_scratch_size_opt
+#define esp_nn_set_softmax_scratch_buf esp_nn_set_softmax_scratch_buf_opt
+#define esp_nn_softmax_s8 esp_nn_softmax_s8_opt
--- a/code/components/esp-nn/include/esp_nn_esp32s3.h
+++ b/code/components/esp-nn/include/esp_nn_esp32s3.h
@@ -0,0 +1,261 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+/**
+ * @file        Header definitions to include for esp_nn optimized functions for
+ *              the ESP32-S3 platform
+ */
+
+#pragma once
+
+#include <stdint.h>
+#include "esp_nn_ansi_headers.h"
+
+/************************** Basic math functions *****************************/
+
+
+/**
+ * @brief       elementwise addition
+ *
+ * @note        inputs type: int8_t, output: int8_t
+ *              input offsets: although int32_t, they are contained in 8 bits [-128, 127]
+ *
+ *              shift values are expected to be <= 0
+ */
+void esp_nn_add_elementwise_s8_esp32s3(const int8_t *input1_data,
+                                       const int8_t *input2_data,
+                                       const int32_t input1_offset,
+                                       const int32_t input2_offset,
+                                       const int32_t input1_mult,
+                                       const int32_t input2_mult,
+                                       const int32_t input1_shift,
+                                       const int32_t input2_shift,
+                                       const int32_t left_shift,
+                                       int8_t *output,
+                                       const int32_t out_offset,
+                                       const int32_t out_mult,
+                                       const int32_t out_shift,
+                                       const int32_t activation_min,
+                                       const int32_t activation_max,
+                                       const int32_t size);
+
+/**
+ * @brief       elementwise multiplication
+ *
+ * @note        inputs type: int8_t, output: int8_t
+ *              input offsets: although int32_t, they are contained in 8 bits [-128, 127]
+ *
+ *              output shift is expected to be <= 0
+ */
+void esp_nn_mul_elementwise_s8_esp32s3(const int8_t *input1_data,
+                                       const int8_t *input2_data,
+                                       const int32_t input1_offset,
+                                       const int32_t input2_offset,
+                                       int8_t *output,
+                                       const int32_t out_offset,
+                                       const int32_t out_mult,
+                                       const int32_t out_shift,
+                                       const int32_t activation_min,
+                                       const int32_t activation_max,
+                                       const int32_t size);
+
+
+/************************** Convolution functions *****************************/
+
+/**
+ * @brief       depthwise convolution per channel
+ *
+ * @note        inputs type: int8_t, output: int8_t
+ *              Version used in tflite is per channel.
+ *              This version follows the same footsprints.
+ *              Meaning, it has per out_channel shift and multiplier for
+ *              requantization
+ *
+ *              optimization notes: Though input_offset is int32 type,
+ *              offset values are contained in 8 bits [-128, 127]
+ */
+void esp_nn_depthwise_conv_s8_esp32s3(const int8_t *input_data,
+                                      const uint16_t input_wd,
+                                      const uint16_t input_ht,
+                                      const uint16_t channels,
+                                      const int32_t input_offset,
+                                      const uint16_t pad_wd,
+                                      const uint16_t pad_ht,
+                                      const uint16_t stride_wd,
+                                      const uint16_t stride_ht,
+                                      const uint16_t ch_mult,
+                                      const int8_t *filter_data,
+                                      const uint16_t filter_wd,
+                                      const uint16_t filter_ht,
+                                      const int32_t *bias,
+                                      int8_t *out_data,
+                                      const uint16_t out_wd,
+                                      const uint16_t out_ht,
+                                      const int32_t out_offset,
+                                      const int32_t *out_shift,
+                                      const int32_t *out_mult,
+                                      const int32_t activation_min,
+                                      const int32_t activation_max);
+
+/**
+ * @brief       2d - convolution channelwise
+ *
+ * @note        operation: result += (input + offset) * filter
+ *
+ *              inputs type: int8_t, output: int8_t
+ *              input offsets: although int32_t, they are contained in 8 bits [-128, 127]
+ */
+void esp_nn_conv_s8_esp32s3(const int8_t *input_data,
+                            const uint16_t input_wd,
+                            const uint16_t input_ht,
+                            const uint16_t in_channels,
+                            const int32_t input_offset,
+                            const uint16_t pad_wd,
+                            const uint16_t pad_ht,
+                            const uint16_t stride_wd,
+                            const uint16_t stride_ht,
+                            const int8_t *filter_data,
+                            const uint16_t filter_wd,
+                            const uint16_t filter_ht,
+                            const int32_t *bias,
+                            int8_t *out_data,
+                            const uint16_t out_wd,
+                            const uint16_t out_ht,
+                            const uint16_t out_channels,
+                            const int32_t out_offset,
+                            const int32_t *out_shift,
+                            const int32_t *out_mult,
+                            const int32_t activation_min,
+                            const int32_t activation_max);
+
+int esp_nn_get_conv_scratch_size_esp32s3(const uint16_t input_wd,
+                                         const uint16_t input_ht,
+                                         const uint16_t in_ch,
+                                         const uint16_t out_ch,
+                                         const uint16_t filter_wd,
+                                         const uint16_t filter_ht);
+void esp_nn_set_conv_scratch_buf_esp32s3(const void *buf);
+
+int esp_nn_get_depthwise_conv_scratch_size_esp32s3(const uint16_t input_wd,
+                                                   const uint16_t input_ht,
+                                                   const uint16_t channels,
+                                                   const uint16_t ch_mult,
+                                                   const uint16_t filter_wd,
+                                                   const uint16_t filter_ht);
+void esp_nn_set_depthwise_conv_scratch_buf_esp32s3(const void *buf);
+
+/************************** Pooling functions *****************************/
+
+/**
+ * @brief       max_pool
+ *
+ * @note        inputs type: int8_t, output: int8_t
+ *              input offsets: although int32_t, they are contained in 8 bits [-128, 127]
+ */
+void esp_nn_max_pool_s8_esp32s3(const int8_t *input,
+                                const uint16_t input_wd,
+                                const uint16_t input_ht,
+                                int8_t *output,
+                                const uint16_t output_wd,
+                                const uint16_t output_ht,
+                                const uint16_t stride_wd,
+                                const uint16_t stride_ht,
+                                const uint16_t filter_wd,
+                                const uint16_t filter_ht,
+                                const uint16_t pad_wd,
+                                const uint16_t pad_ht,
+                                const int32_t activation_min,
+                                const int32_t activation_max,
+                                const uint16_t channels);
+
+/**
+ * @brief       avg_pool
+ *
+ * @note        inputs type: int8_t, output: int8_t
+ *              input offsets: although int32_t, they are contained in 8 bits [-128, 127]
+ */
+void esp_nn_avg_pool_s8_esp32s3(const int8_t *input,
+                                const uint16_t input_wd,
+                                const uint16_t input_ht,
+                                int8_t *output,
+                                const uint16_t output_wd,
+                                const uint16_t output_ht,
+                                const uint16_t stride_wd,
+                                const uint16_t stride_ht,
+                                const uint16_t filter_wd,
+                                const uint16_t filter_ht,
+                                const uint16_t pad_wd,
+                                const uint16_t pad_ht,
+                                const int32_t activation_min,
+                                const int32_t activation_max,
+                                const uint16_t channels);
+
+
+/************************** Fully connected functions *****************************/
+
+/**
+ * @brief       fully connected
+ *
+ * @note        inputs type: int8_t, output: int8_t
+ *              input offsets: although int32_t, they are contained in 8 bits [-128, 127]
+ *
+ *              Current version works only on aligned input.
+ *              row_len and channels should both be multiple of 8.
+ */
+void esp_nn_fully_connected_s8_esp32s3(const int8_t *input_data,
+                                       const int32_t input_offset,
+                                       const uint16_t row_len,
+                                       const int8_t *filter_data,
+                                       const int32_t filter_offset,
+                                       const int32_t *bias,
+                                       int8_t *out_data,
+                                       const uint16_t out_channels,
+                                       const int32_t out_offset,
+                                       const int32_t out_shift,
+                                       const int32_t out_mult,
+                                       const int32_t activation_min,
+                                       const int32_t activation_max);
+
+/**
+ * @brief       relu6
+ *
+ * @note        inout: int8_t
+ */
+void esp_nn_relu6_s8_esp32s3(int8_t *data, uint16_t size);
+
+/********************** function defines ***************************/
+
+#define esp_nn_add_elementwise_s8 esp_nn_add_elementwise_s8_esp32s3
+#define esp_nn_mul_elementwise_s8 esp_nn_mul_elementwise_s8_esp32s3
+
+#define esp_nn_depthwise_conv_s8 esp_nn_depthwise_conv_s8_esp32s3
+
+#define esp_nn_get_conv_scratch_size esp_nn_get_conv_scratch_size_esp32s3
+#define esp_nn_set_conv_scratch_buf esp_nn_set_conv_scratch_buf_esp32s3
+
+#define esp_nn_get_depthwise_conv_scratch_size esp_nn_get_depthwise_conv_scratch_size_esp32s3
+#define esp_nn_set_depthwise_conv_scratch_buf esp_nn_set_depthwise_conv_scratch_buf_esp32s3
+
+#define esp_nn_conv_s8 esp_nn_conv_s8_esp32s3
+
+#define esp_nn_relu6_s8 esp_nn_relu6_s8_esp32s3
+
+#define esp_nn_avg_pool_s8 esp_nn_avg_pool_s8_esp32s3
+#define esp_nn_max_pool_s8 esp_nn_max_pool_s8_esp32s3
+
+#define esp_nn_fully_connected_s8 esp_nn_fully_connected_s8_esp32s3
+
+#define esp_nn_get_softmax_scratch_size esp_nn_get_softmax_scratch_size_opt
+#define esp_nn_set_softmax_scratch_buf esp_nn_set_softmax_scratch_buf_opt
+#define esp_nn_softmax_s8 esp_nn_softmax_s8_opt
--- a/code/components/esp-nn/src/activation_functions/esp_nn_relu_ansi.c
+++ b/code/components/esp-nn/src/activation_functions/esp_nn_relu_ansi.c
@@ -0,0 +1,30 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+#include <stdlib.h>
+
+#include <common_functions.h>
+
+void esp_nn_relu6_s8_ansi(int8_t *data, uint16_t size)
+{
+    int32_t i;
+
+    for (i = 0; i < size; i++) {
+        int32_t ip = data[i];
+
+        ip = max(ip, 0);
+        data[i] = min(ip, 6);
+    }
+}
--- a/code/components/esp-nn/src/basic_math/esp_nn_add_ansi.c
+++ b/code/components/esp-nn/src/basic_math/esp_nn_add_ansi.c
@@ -0,0 +1,97 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+
+#include <common_functions.h>
+
+void esp_nn_add_elementwise_u8_ansi(const uint8_t *input1_data,
+                                    const uint8_t *input2_data,
+                                    const int32_t input1_offset,
+                                    const int32_t input2_offset,
+                                    const int32_t input1_mult,
+                                    const int32_t input2_mult,
+                                    const int32_t input1_shift,
+                                    const int32_t input2_shift,
+                                    const int32_t left_shift,
+                                    uint8_t *output,
+                                    const int32_t out_offset,
+                                    const int32_t out_mult,
+                                    const int32_t out_shift,
+                                    const int32_t activation_min,
+                                    const int32_t activation_max,
+                                    const int32_t size)
+{
+    for (int i = 0; i < size; i++) {
+        int32_t tmp1 = input1_data[i] + input1_offset;
+        int32_t tmp2 = input2_data[i] + input2_offset;
+
+        tmp1 <<= left_shift;
+        tmp2 <<= left_shift;
+
+        tmp1 = esp_nn_sat_round_doubling_high_mul(tmp1, input1_mult);
+        tmp2 = esp_nn_sat_round_doubling_high_mul(tmp2, input2_mult);
+
+        tmp1 = esp_nn_div_by_power_of_two(tmp1, -input1_shift);
+        tmp2 = esp_nn_div_by_power_of_two(tmp2, -input2_shift);
+
+        int32_t out = tmp1 + tmp2;
+        out = esp_nn_sat_round_doubling_high_mul(out, out_mult);
+        out = esp_nn_div_by_power_of_two(out, -out_shift);
+        out = out + out_offset;
+
+        out = max(activation_min, min(out, activation_max));
+        output[i] = (uint8_t) out;
+    }
+}
+
+void esp_nn_add_elementwise_s8_ansi(const int8_t *input1_data,
+                                    const int8_t *input2_data,
+                                    const int32_t input1_offset,
+                                    const int32_t input2_offset,
+                                    const int32_t input1_mult,
+                                    const int32_t input2_mult,
+                                    const int32_t input1_shift,
+                                    const int32_t input2_shift,
+                                    const int32_t left_shift,
+                                    int8_t *output,
+                                    const int32_t out_offset,
+                                    const int32_t out_mult,
+                                    const int32_t out_shift,
+                                    const int32_t activation_min,
+                                    const int32_t activation_max,
+                                    const int32_t size)
+{
+    for (int i = 0; i < size; i++) {
+        int32_t tmp1 = input1_data[i] + input1_offset;
+        int32_t tmp2 = input2_data[i] + input2_offset;
+
+        tmp1 <<= left_shift;
+        tmp2 <<= left_shift;
+
+        tmp1 = esp_nn_sat_round_doubling_high_mul(tmp1, input1_mult);
+        tmp2 = esp_nn_sat_round_doubling_high_mul(tmp2, input2_mult);
+
+        tmp1 = esp_nn_div_by_power_of_two(tmp1, -input1_shift);
+        tmp2 = esp_nn_div_by_power_of_two(tmp2, -input2_shift);
+
+        int32_t out = tmp1 + tmp2;
+        out = esp_nn_sat_round_doubling_high_mul(out, out_mult);
+        out = esp_nn_div_by_power_of_two(out, -out_shift);
+        out = out + out_offset;
+
+        out = max(activation_min, min(out, activation_max));
+        output[i] = (int8_t) out;
+    }
+}
--- a/code/components/esp-nn/src/basic_math/esp_nn_mul_ansi.c
+++ b/code/components/esp-nn/src/basic_math/esp_nn_mul_ansi.c
@@ -0,0 +1,42 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+
+#include <common_functions.h>
+
+void esp_nn_mul_elementwise_s8_ansi(const int8_t *input1_data,
+                                    const int8_t *input2_data,
+                                    const int32_t input1_offset,
+                                    const int32_t input2_offset,
+                                    int8_t *output,
+                                    const int32_t out_offset,
+                                    const int32_t out_mult,
+                                    const int32_t out_shift,
+                                    const int32_t activation_min,
+                                    const int32_t activation_max,
+                                    const int32_t size)
+{
+    for (int i = 0; i < size; i++) {
+        int32_t tmp1 = input1_data[i] + input1_offset;
+        int32_t tmp2 = input2_data[i] + input2_offset;
+
+        int32_t out = tmp1 * tmp2;
+        out = esp_nn_multiply_by_quantized_mult(out, out_mult, out_shift);
+        out = out + out_offset;
+
+        out = max(activation_min, min(out, activation_max));
+        output[i] = (int8_t) out;
+    }
+}
--- a/code/components/esp-nn/src/common/common_functions.h
+++ b/code/components/esp-nn/src/common/common_functions.h
@@ -0,0 +1,218 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include <stdint.h>
+#include <stdbool.h>
+#include <string.h>
+
+/**
+ * c99 standard still doesn't strictly inline functions
+ * We need to use attribute as well to do this.
+ */
+#define __NN_FORCE_INLINE__ __attribute((always_inline)) static inline
+
+/* min/max macros */
+#ifndef max
+#define max(a, b) ({            \
+    __typeof__ (a) _a = (a);    \
+    __typeof__ (b) _b = (b);    \
+    _a > _b ? _a : _b;          \
+})
+
+#define min(a, b) ({            \
+    __typeof__ (a) _a = (a);    \
+    __typeof__ (b) _b = (b);    \
+    _a < _b ? _a : _b;          \
+})
+#endif
+
+__NN_FORCE_INLINE__ int32_t esp_nn_clz32(uint32_t in)
+{
+    __asm__ volatile("nsau %0, %0" : "+r" (in));
+    return in;
+}
+
+__NN_FORCE_INLINE__ int32_t esp_nn_pick_sat_high32_of64(int64_t val64)
+{
+    int32_t sign = (int32_t) (val64 >> 63);
+    int32_t to_add = sign & ((1ul << 31) - 1);
+    return (int32_t) ((int64_t) (val64 + to_add) >> 31);
+}
+
+/**
+ * Signed saturate a 32 bit value to 8 bits keeping output in 32 bit variable.
+ */
+__NN_FORCE_INLINE__ int32_t esp_nn_saturate8(int32_t in)
+{
+    __asm__ volatile("clamps %0, %0, 7" : "+a"(in));
+    return in;
+}
+
+__NN_FORCE_INLINE__ int32_t esp_nn_sat_round_doubling_high_mul(int32_t in0, int32_t in1)
+{
+    int32_t result;
+    int64_t in0_64 = (int64_t) in0;
+    bool overflow = (in0 == in1) && (in0 == (int32_t) INT32_MIN);
+
+    /* Nudge value */
+    int64_t nudge_val = 1 << 30;
+    if ((in0 < 0) ^ (in1 < 0)) {
+        nudge_val = 1 - nudge_val;
+    }
+
+    /* Multiply and add nudge */
+    int64_t mult = in0_64 * in1 + nudge_val;
+
+    /* Round and pickup 32 bits */
+    result = esp_nn_pick_sat_high32_of64(mult);
+
+    return overflow ? INT32_MAX : result;
+}
+
+/**
+ * fast version
+ * this will fail for values closer to INT32_MAX and INT32_MIN by `1 << (exponent - 1)`.
+ * We can afford to do this because we are at the very last stage of filter.
+ * Also it is pretty rare condition as our output is going to be 8 bit.
+ */
+__NN_FORCE_INLINE__ int32_t esp_nn_div_by_power_of_two_fast(int32_t val, int32_t exponent)
+{
+    int32_t to_add = (1 << (exponent - 1)) - (val < 0);
+    return (int32_t) ((val + to_add) >> exponent);
+}
+
+__NN_FORCE_INLINE__ int32_t esp_nn_div_by_power_of_two(int32_t val, int32_t exponent)
+{
+    int32_t result;
+
+    const int32_t mask = (1 << exponent) - 1;
+    const int32_t remainder = val & mask;
+
+    result = val >> exponent;
+    int32_t threshold = (mask >> 1) + (result < 0);
+
+    if (remainder > threshold) {
+        result += 1;
+    }
+    return result;
+}
+
+__NN_FORCE_INLINE__ int32_t esp_nn_multiply_by_quantized_mult(int32_t x, int32_t mult, int32_t shift)
+{
+    int32_t left_shift = shift > 0 ? shift : 0;
+    int32_t right_shift = shift > 0 ? 0 : -shift;
+    int32_t result = esp_nn_sat_round_doubling_high_mul(x * (1 << left_shift), mult);
+    return esp_nn_div_by_power_of_two(result, right_shift);
+}
+
+__NN_FORCE_INLINE__ int32_t esp_nn_multiply_by_quantized_mult_fast(int32_t x, int32_t mult, int32_t shift)
+{
+    int32_t left_shift = max(shift, 0);
+    int32_t right_shift = left_shift - shift;
+
+    int64_t nudge_val = 1 << 30;
+    int64_t in0_64 = (int64_t) (x << left_shift);
+
+    /* Multiply and add nudge */
+    int64_t mult_64 = in0_64 * mult + nudge_val;
+    int32_t result = (int32_t) (mult_64 >> 31);
+    if (right_shift) {
+        result = esp_nn_div_by_power_of_two_fast(result, right_shift);
+    }
+    return result;
+}
+
+static void esp_nn_aligned_s8_pad_with_value(const int8_t *src, int8_t *dst,
+                                             const uint16_t input_wd,
+                                             const uint16_t input_ht,
+                                             const uint16_t channels,
+                                             const int32_t pad_val,
+                                             const uint16_t pad_wd,
+                                             const uint16_t pad_ht)
+{
+    /* memset with pad_val */
+    memset(dst, pad_val, ((input_wd + 2 * pad_wd) * (input_ht + 2 * pad_ht)) * channels * 2);
+    dst += (pad_wd + input_wd + pad_wd) * channels;
+
+    for (int i = 0; i < input_ht; i++) {
+        dst += pad_wd * channels;
+        for (int j = 0; j < input_wd * channels; j++) {
+            *dst++ = *src++;
+        }
+        dst += pad_wd * channels;
+    }
+}
+
+#if 0
+static void esp_nn_aligned_s8_pad_end_with_value(const int8_t *src, int8_t *dst,
+                                                 const uint16_t input_wd,
+                                                 const uint16_t input_ht,
+                                                 const uint16_t channels,
+                                                 const int32_t pad_val,
+                                                 const uint16_t pad_wd,
+                                                 const uint16_t pad_ht)
+{
+    for (int i = 0; i < input_ht; i++) {
+        for (int j = 0; j < input_wd * channels; j++) {
+            *dst++ = *src++;
+        }
+        memset(dst, pad_val, pad_wd * channels);
+        dst += pad_wd * channels;
+    }
+    /* pad end `pad_ht` lines at end */
+    memset(dst, pad_val, (input_wd + pad_wd) * pad_ht * channels);
+}
+#endif
+
+/**
+ * @brief       convert 8 bit input data to 16 bit
+ *
+ * @param       src int8_t source data
+ * @param       dst int16_t dst data
+ * @param       size length of data
+ * @param       offset  offset to be added to src data. Range: [-128, 127]
+ */
+__NN_FORCE_INLINE__ void esp_nn_s8_to_s16_with_offset(const int8_t *src, int16_t *dst,
+                                                      const int size, const int32_t offset)
+{
+    int i = 0;
+    for (; i < size; i += 2) {
+        dst[i + 0] = src[i + 0] + offset;
+        dst[i + 1] = src[i + 1] + offset;
+    }
+    if(i < size) {
+        dst[i] = src[i] + offset;
+    }
+}
+
+/**
+ * @brief       convert 8 bit input data to 16 bit
+ *
+ * @param       src int8_t source data
+ * @param       dst int16_t dst data
+ * @param       size length of data
+ */
+__NN_FORCE_INLINE__ void esp_nn_s8_to_s16(const int8_t *src, int16_t *dst, const int size)
+{
+    int i = 0;
+    for (; i < size; i += 2) {
+        dst[i + 0] = src[i + 0];
+        dst[i + 1] = src[i + 1];
+    }
+    if(i < size) {
+        dst[i] = src[i];
+    }
+}
--- a/code/components/esp-nn/src/convolution/esp_nn_conv_ansi.c
+++ b/code/components/esp-nn/src/convolution/esp_nn_conv_ansi.c
@@ -0,0 +1,175 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+
+#include <common_functions.h>
+
+int esp_nn_get_conv_scratch_size_ansi(const uint16_t input_wd,
+                                      const uint16_t input_ht,
+                                      const uint16_t in_ch,
+                                      const uint16_t out_ch,
+                                      const uint16_t filter_wd,
+                                      const uint16_t filter_ht)
+{
+    return 0;
+}
+
+void esp_nn_set_conv_scratch_buf_ansi(const void *buf)
+{
+
+}
+
+/**
+ * Assumption 1: i/p channels == o/p channels
+ * Assumption 2: Pointers are valid
+ * Assumption 3: dialation width = 1
+ */
+void esp_nn_conv_u8_ansi(const uint8_t *input_data,
+                         const uint16_t input_wd,
+                         const uint16_t input_ht,
+                         const uint16_t in_channels,
+                         const int32_t input_offset,
+                         const uint16_t pad_wd,
+                         const uint16_t pad_ht,
+                         const uint16_t stride_wd,
+                         const uint16_t stride_ht,
+                         const uint8_t *filter_data,
+                         const uint16_t filter_wd,
+                         const uint16_t filter_ht,
+                         const int32_t filter_offset,
+                         const int32_t *bias,
+                         uint8_t *out_data,
+                         const uint16_t out_wd,
+                         const uint16_t out_ht,
+                         const uint16_t out_channels,
+                         const int32_t out_offset,
+                         const int32_t out_shift,
+                         const int32_t out_mult,
+                         const int32_t activation_min,
+                         const int32_t activation_max)
+{
+    for (int out_y = 0; out_y < out_ht; out_y++) { //height loop
+        const int16_t base_y = (out_y * stride_ht) - pad_ht;
+        for (int out_x = 0; out_x < out_wd; out_x++) { //width_loop
+            const int16_t base_x = (out_x * stride_wd) - pad_wd;
+            for (int out_ch_idx = 0; out_ch_idx < out_channels; out_ch_idx++) {//channel_loop
+                int32_t result = 0;
+
+                /* Select filter so as the point doesn't lie outside block */
+                int filter_y_start = max(0, -base_y);
+                int filter_x_start = max(0, -base_x);
+                int filter_y_end = min(filter_ht, input_ht - base_y);
+                int filter_x_end = min(filter_wd, input_wd - base_x);
+
+                for (int filter_y_idx = filter_y_start; filter_y_idx < filter_y_end; filter_y_idx++) {
+                    const int32_t idx_y = base_y + filter_y_idx;
+                    for (int filter_x_idx = filter_x_start; filter_x_idx < filter_x_end; filter_x_idx++) {
+                        const int32_t idx_x = base_x + filter_x_idx;
+                        for (int in_ch_idx = 0; in_ch_idx < in_channels; in_ch_idx++) {
+                            int32_t input_index = (idx_y * input_wd + idx_x) * in_channels + in_ch_idx;
+                            int32_t filter_index = ((out_ch_idx * filter_ht + filter_y_idx)
+                                                    * filter_wd + filter_x_idx) * in_channels
+                                                   + in_ch_idx;
+                            int32_t input_val = input_data[input_index] + input_offset;
+                            int32_t filter_val = filter_data[filter_index] + filter_offset;
+                            result += input_val * filter_val;
+                        }
+                    }
+                }
+                if (bias) {
+                    result += bias[out_ch_idx];
+                }
+                result = esp_nn_multiply_by_quantized_mult(result, out_mult, out_shift);
+                result += out_offset;
+                result = max(result, activation_min);
+                result = min(result, activation_max);
+
+                int out_index = (out_y * out_wd + out_x) * out_channels + out_ch_idx;
+                out_data[out_index] = (uint8_t) result;
+            }
+        }
+    }
+}
+
+/**
+ * Assumption 1: i/p channels == o/p channels
+ * Assumption 2: Pointers are valid
+ * Assumption 3: dialation width = 1
+ */
+void esp_nn_conv_s8_ansi(const int8_t *input_data,
+                         const uint16_t input_wd,
+                         const uint16_t input_ht,
+                         const uint16_t in_channels,
+                         const int32_t input_offset,
+                         const uint16_t pad_wd,
+                         const uint16_t pad_ht,
+                         const uint16_t stride_wd,
+                         const uint16_t stride_ht,
+                         const int8_t *filter_data,
+                         const uint16_t filter_wd,
+                         const uint16_t filter_ht,
+                         const int32_t *bias,
+                         int8_t *out_data,
+                         const uint16_t out_wd,
+                         const uint16_t out_ht,
+                         const uint16_t out_channels,
+                         const int32_t out_offset,
+                         const int32_t *out_shift,
+                         const int32_t *out_mult,
+                         const int32_t activation_min,
+                         const int32_t activation_max)
+{
+    int32_t out_ch_idx, out_y, out_x, in_ch_idx, filter_y_idx, filter_x_idx;
+
+    for (out_y = 0; out_y < out_ht; out_y++) {
+        for (out_x = 0; out_x < out_wd; out_x++) {
+            for (out_ch_idx = 0; out_ch_idx < out_channels; out_ch_idx++) {
+                int32_t conv_out = 0;
+
+                const int32_t base_y = stride_ht * out_y - pad_ht;
+                const int32_t base_x = stride_wd * out_x - pad_wd;
+
+                const int32_t filter_y_start = max(0, -base_y);
+                const int32_t filter_x_start = max(0, -base_x);
+
+                const int32_t filter_y_end = min(filter_ht, input_ht - base_y);
+                const int32_t filter_x_end = min(filter_wd, input_wd - base_x);
+
+                for (filter_y_idx = filter_y_start; filter_y_idx < filter_y_end; filter_y_idx++) {
+                    for (filter_x_idx = filter_x_start; filter_x_idx < filter_x_end; filter_x_idx++) {
+                        const int32_t in_row = base_y + filter_y_idx;
+                        const int32_t in_col = base_x + filter_x_idx;
+                        int32_t input_base_offset = (in_row * input_wd + in_col) * in_channels;
+                        int32_t filter_base_offset = out_ch_idx * in_channels * filter_ht * filter_wd +
+                                                       (filter_y_idx * filter_wd + filter_x_idx) * in_channels;
+                        for (in_ch_idx = 0; in_ch_idx < in_channels; in_ch_idx++) {
+                            conv_out +=
+                                (input_data[input_base_offset + in_ch_idx] + input_offset) *
+                                filter_data[filter_base_offset + in_ch_idx];
+                        }
+                    }
+                }
+                if (bias) {
+                    conv_out += bias[out_ch_idx];
+                }
+                conv_out = esp_nn_multiply_by_quantized_mult(conv_out, out_mult[out_ch_idx], out_shift[out_ch_idx]);
+                conv_out += out_offset;
+                conv_out = max(conv_out, activation_min);
+                conv_out = min(conv_out, activation_max);
+                *out_data++ = (int8_t) conv_out;
+            }
+        }
+    }
+}
--- a/code/components/esp-nn/src/convolution/esp_nn_conv_esp32s3.c
+++ b/code/components/esp-nn/src/convolution/esp_nn_conv_esp32s3.c
@@ -0,0 +1,436 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+#include <stdio.h>
+
+#include <common_functions.h>
+
+static int16_t *scratch_buffer = NULL;
+
+extern void esp_nn_conv_s16_mult8_1x1_esp32s3(const int8_t *input_data,
+                                              const uint16_t input_wd,
+                                              const uint16_t input_ht,
+                                              const uint16_t in_channels,
+                                              const int32_t input_offset,
+                                              const int16_t *filter_data,
+                                              const int32_t *bias,
+                                              int8_t *out_data,
+                                              const uint16_t out_wd,
+                                              const uint16_t out_ht,
+                                              const uint16_t out_channels,
+                                              const int32_t out_offset,
+                                              const int32_t *out_shift,
+                                              const int32_t *out_mult,
+                                              const int32_t activation_min,
+                                              const int32_t activation_max,
+                                              void *buffer /* scratch buffer */);
+
+extern void esp_nn_conv_s16_mult4_1x1_esp32s3(const int16_t *input_data,
+                                              const uint16_t input_wd,
+                                              const uint16_t input_ht,
+                                              const uint16_t in_channels,
+                                              const int16_t *filter_data,
+                                              const int32_t *bias,
+                                              int8_t *out_data,
+                                              const uint16_t out_wd,
+                                              const uint16_t out_ht,
+                                              const uint16_t out_channels,
+                                              const int32_t out_offset,
+                                              const int32_t *out_shift,
+                                              const int32_t *out_mult,
+                                              const int32_t activation_min,
+                                              const int32_t activation_max,
+                                              void *buffer /* scratch buffer */);
+
+extern void esp_nn_conv_s16_mult8_esp32s3(const int16_t *input_data,
+                                          const uint16_t input_wd,
+                                          const uint16_t input_ht,
+                                          const uint16_t in_channels,
+                                          const uint16_t pad_wd,
+                                          const uint16_t pad_ht,
+                                          const uint16_t stride_wd,
+                                          const uint16_t stride_ht,
+                                          const int16_t *filter_data,
+                                          const uint16_t filter_wd,
+                                          const uint16_t filter_ht,
+                                          const int32_t *bias,
+                                          int8_t *out_data,
+                                          const uint16_t out_wd,
+                                          const uint16_t out_ht,
+                                          const uint16_t out_channels,
+                                          const int32_t out_offset,
+                                          const int32_t *out_shift,
+                                          const int32_t *out_mult,
+                                          const int32_t activation_min,
+                                          const int32_t activation_max);
+
+extern void esp_nn_aligned_s8_to_s16_with_offset_esp32s3(const int8_t *src, int16_t *dst,
+                                                         const int size, const int32_t offset);
+
+extern void esp_nn_s8_to_s16_esp32s3(const int8_t *src, int16_t *dst, const int size);
+
+static void esp_nn_conv_s8_unrolled(const int8_t *input_data,
+                                    const uint16_t input_wd,
+                                    const uint16_t input_ht,
+                                    const uint16_t in_channels,
+                                    const int32_t input_offset,
+                                    const uint16_t pad_wd,
+                                    const uint16_t pad_ht,
+                                    const uint16_t stride_wd,
+                                    const uint16_t stride_ht,
+                                    const int8_t *filter_data,
+                                    const uint16_t filter_wd,
+                                    const uint16_t filter_ht,
+                                    const int32_t *bias,
+                                    int8_t *out_data,
+                                    const uint16_t out_wd,
+                                    const uint16_t out_ht,
+                                    const uint16_t out_channels,
+                                    const int32_t out_offset,
+                                    const int32_t *out_shift,
+                                    const int32_t *out_mult,
+                                    const int32_t activation_min,
+                                    const int32_t activation_max)
+{
+    int32_t out_ch_idx, out_y, out_x, in_ch_idx, filter_y_idx, filter_x_idx;
+
+    for (out_y = 0; out_y < out_ht; out_y++) {
+        for (out_x = 0; out_x < out_wd; out_x++) {
+            for (out_ch_idx = 0; out_ch_idx < out_channels; out_ch_idx++) {
+                int32_t conv_out = 0;
+
+                const int32_t base_y = stride_ht * out_y - pad_ht;
+                const int32_t base_x = stride_wd * out_x - pad_wd;
+
+                const int32_t filter_y_start = max(0, -base_y);
+                const int32_t filter_x_start = max(0, -base_x);
+
+                const int32_t filter_y_end = min(filter_ht, input_ht - base_y);
+                const int32_t filter_x_end = min(filter_wd, input_wd - base_x);
+
+                for (filter_y_idx = filter_y_start; filter_y_idx < filter_y_end; filter_y_idx++) {
+                    for (filter_x_idx = filter_x_start; filter_x_idx < filter_x_end; filter_x_idx++) {
+                        const int32_t in_row = base_y + filter_y_idx;
+                        const int32_t in_col = base_x + filter_x_idx;
+                        int32_t input_base_offset = (in_row * input_wd + in_col) * in_channels;
+                        int32_t filter_base_offset = out_ch_idx * in_channels * filter_ht * filter_wd +
+                                                       (filter_y_idx * filter_wd + filter_x_idx) * in_channels;
+                        for (in_ch_idx = 0; in_ch_idx < in_channels; in_ch_idx++) {
+                            conv_out +=
+                                (input_data[input_base_offset + in_ch_idx] + input_offset) *
+                                filter_data[filter_base_offset + in_ch_idx];
+                        }
+                    }
+                }
+                if (bias) {
+                    conv_out += bias[out_ch_idx];
+                }
+                conv_out = esp_nn_multiply_by_quantized_mult_fast(conv_out, out_mult[out_ch_idx], out_shift[out_ch_idx]);
+                conv_out += out_offset;
+                conv_out = max(conv_out, activation_min);
+                conv_out = min(conv_out, activation_max);
+                *out_data++ = (int8_t) conv_out;
+            }
+        }
+    }
+}
+
+static void esp_nn_conv_s8_pad_valid(const int8_t *input_data,
+                                     const uint16_t input_wd,
+                                     const uint16_t input_ht,
+                                     const uint16_t in_channels,
+                                     const int32_t input_offset,
+                                     const uint16_t stride_wd,
+                                     const uint16_t stride_ht,
+                                     const int8_t *filter_data,
+                                     const uint16_t filter_wd,
+                                     const uint16_t filter_ht,
+                                     const int32_t *bias,
+                                     int8_t *out_data,
+                                     const uint16_t out_wd,
+                                     const uint16_t out_ht,
+                                     const uint16_t out_channels,
+                                     const int32_t out_offset,
+                                     const int32_t *out_shift,
+                                     const int32_t *out_mult,
+                                     const int32_t activation_min,
+                                     const int32_t activation_max)
+{
+    int32_t out_ch_idx, out_y, out_x, in_ch_idx, filter_y_idx, filter_x_idx;
+
+    for (out_y = 0; out_y < out_ht; out_y++) {
+        for (out_x = 0; out_x < out_wd; out_x++) {
+            for (out_ch_idx = 0; out_ch_idx < out_channels; out_ch_idx++) {
+                int32_t conv_out = 0;
+
+                const int32_t base_y = stride_ht * out_y;
+                const int32_t base_x = stride_wd * out_x;
+
+                for (filter_y_idx = 0; filter_y_idx < filter_ht; filter_y_idx++) {
+                    for (filter_x_idx = 0; filter_x_idx < filter_wd; filter_x_idx++) {
+                        const int32_t in_row = base_y + filter_y_idx;
+                        const int32_t in_col = base_x + filter_x_idx;
+                        int32_t input_base_offset = (in_row * input_wd + in_col) * in_channels;
+                        int32_t filter_base_offset = out_ch_idx * in_channels * filter_ht * filter_wd +
+                                                       (filter_y_idx * filter_wd + filter_x_idx) * in_channels;
+                        const int8_t *input_data_ptr = input_data + input_base_offset;
+                        const int8_t *filter_data_ptr = filter_data + filter_base_offset;
+                        for (in_ch_idx = 0; in_ch_idx < in_channels; in_ch_idx++) {
+                            conv_out += (*input_data_ptr++ + input_offset) * *filter_data_ptr++;
+                        }
+                    }
+                }
+                if (bias) {
+                    conv_out += bias[out_ch_idx];
+                }
+                conv_out = esp_nn_multiply_by_quantized_mult_fast(conv_out, out_mult[out_ch_idx], out_shift[out_ch_idx]);
+                conv_out += out_offset;
+                conv_out = max(conv_out, activation_min);
+                conv_out = min(conv_out, activation_max);
+                *out_data++ = (int8_t) conv_out;
+            }
+        }
+    }
+}
+
+static void esp_nn_conv_s8_pad_valid_3x3(const int8_t *input_data,
+                                         const uint16_t input_wd,
+                                         const uint16_t input_ht,
+                                         const uint16_t in_channels,
+                                         const int32_t input_offset,
+                                         const uint16_t stride_wd,
+                                         const uint16_t stride_ht,
+                                         const int8_t *filter_data,
+                                         const int32_t *bias,
+                                         int8_t *out_data,
+                                         const uint16_t out_wd,
+                                         const uint16_t out_ht,
+                                         const uint16_t out_channels,
+                                         const int32_t out_offset,
+                                         const int32_t *out_shift,
+                                         const int32_t *out_mult,
+                                         const int32_t activation_min,
+                                         const int32_t activation_max)
+{
+    int32_t out_ch_idx, out_y, out_x, in_ch_idx, filter_y_idx, filter_x_idx;
+
+    for (out_y = 0; out_y < out_ht; out_y++) {
+        for (out_x = 0; out_x < out_wd; out_x++) {
+            const int32_t base_y = stride_ht * out_y;
+            const int32_t base_x = stride_wd * out_x;
+            for (out_ch_idx = 0; out_ch_idx < out_channels; out_ch_idx++) {
+                int32_t conv_out = 0;
+                for (filter_y_idx = 0; filter_y_idx < 3; filter_y_idx++) {
+                    for (filter_x_idx = 0; filter_x_idx < 3; filter_x_idx++) {
+                        const int32_t in_row = base_y + filter_y_idx;
+                        const int32_t in_col = base_x + filter_x_idx;
+                        int32_t input_base_offset = (in_row * input_wd + in_col) * in_channels;
+                        int32_t filter_base_offset = out_ch_idx * in_channels * 3 * 3 +
+                                                       (filter_y_idx * 3 + filter_x_idx) * in_channels;
+                        const int8_t *input_data_ptr = input_data + input_base_offset;
+                        const int8_t *filter_data_ptr = filter_data + filter_base_offset;
+                        for (in_ch_idx = 0; in_ch_idx < in_channels; in_ch_idx++) {
+                            conv_out += (*input_data_ptr++ + input_offset) * *filter_data_ptr++;
+                        }
+                    }
+                }
+                if (bias) {
+                    conv_out += bias[out_ch_idx];
+                }
+                conv_out = esp_nn_multiply_by_quantized_mult_fast(conv_out, out_mult[out_ch_idx], out_shift[out_ch_idx]);
+                conv_out += out_offset;
+                conv_out = max(conv_out, activation_min);
+                conv_out = min(conv_out, activation_max);
+                *out_data++ = (int8_t) conv_out;
+            }
+        }
+    }
+}
+
+static void esp_nn_conv_s8_pad_valid_ch3_3x3(const int8_t *input_data,
+                                             const uint16_t input_wd,
+                                             const uint16_t input_ht,
+                                             const int32_t input_offset,
+                                             const uint16_t stride_wd,
+                                             const uint16_t stride_ht,
+                                             const int8_t *filter_data,
+                                             const int32_t *bias,
+                                             int8_t *out_data,
+                                             const uint16_t out_wd,
+                                             const uint16_t out_ht,
+                                             const uint16_t out_channels,
+                                             const int32_t out_offset,
+                                             const int32_t *out_shift,
+                                             const int32_t *out_mult,
+                                             const int32_t activation_min,
+                                             const int32_t activation_max)
+{
+    int32_t out_ch_idx, out_y, out_x, filter_y_idx;
+
+    /* use scratch_buffer to pre-compute offset factor */
+    int16_t *filter_sum = (int16_t *) scratch_buffer;
+    const int8_t *filter_ptr = filter_data;
+    for (out_ch_idx = 0; out_ch_idx < out_channels; out_ch_idx++) {
+        int16_t sum_val = 0;
+        for (int i = 0; i < 9; i++) {
+            sum_val += *filter_ptr++;
+            sum_val += *filter_ptr++;
+            sum_val += *filter_ptr++;
+        }
+        *filter_sum++ = sum_val;
+    }
+
+    for (out_y = 0; out_y < out_ht; out_y++) {
+        for (out_x = 0; out_x < out_wd; out_x++) {
+            const int8_t *filter_data_ptr = filter_data;
+            const int32_t base_y = stride_ht * out_y;
+            const int32_t base_x = stride_wd * out_x;
+            const int8_t *input_base_ptr = input_data + (base_y * input_wd + base_x) * 3;
+            int16_t *filter_sum = (int16_t *) scratch_buffer;
+            for (out_ch_idx = 0; out_ch_idx < out_channels; out_ch_idx++) {
+                int32_t conv_out = 0;
+
+                for (filter_y_idx = 0; filter_y_idx < 3; filter_y_idx++) {
+                    const int8_t *input_data_ptr = input_base_ptr + (filter_y_idx * input_wd) * 3;
+                    conv_out += (*input_data_ptr++) * (*filter_data_ptr++);
+                    conv_out += (*input_data_ptr++) * (*filter_data_ptr++);
+                    conv_out += (*input_data_ptr++) * (*filter_data_ptr++);
+
+                    conv_out += (*input_data_ptr++) * (*filter_data_ptr++);
+                    conv_out += (*input_data_ptr++) * (*filter_data_ptr++);
+                    conv_out += (*input_data_ptr++) * (*filter_data_ptr++);
+
+                    conv_out += (*input_data_ptr++) * (*filter_data_ptr++);
+                    conv_out += (*input_data_ptr++) * (*filter_data_ptr++);
+                    conv_out += (*input_data_ptr++) * (*filter_data_ptr++);
+                }
+
+                conv_out += *filter_sum++ * input_offset;
+
+                if (bias) {
+                    conv_out += bias[out_ch_idx];
+                }
+                conv_out = esp_nn_multiply_by_quantized_mult_fast(conv_out, out_mult[out_ch_idx], out_shift[out_ch_idx]);
+                conv_out += out_offset;
+                conv_out = max(conv_out, activation_min);
+                conv_out = min(conv_out, activation_max);
+                *out_data++ = (int8_t) conv_out;
+            }
+        }
+    }
+}
+
+int esp_nn_get_conv_scratch_size_esp32s3(const uint16_t input_wd,
+                                         const uint16_t input_ht,
+                                         const uint16_t in_ch,
+                                         const uint16_t out_ch,
+                                         const uint16_t filter_wd,
+                                         const uint16_t filter_ht)
+{
+    int filter_size = filter_wd * filter_ht * in_ch * out_ch;
+    int input_size = input_wd * input_ht * in_ch;
+    int transpose_buf_size = 8 * in_ch; /* to store intermediate data */
+    int align_buf_size = 32; /* extra buffer for alignment */
+    return 2 * (filter_size + input_size +  transpose_buf_size) + align_buf_size;
+}
+
+void esp_nn_set_conv_scratch_buf_esp32s3(void *buf)
+{
+    scratch_buffer = (int16_t *) buf;
+}
+
+void esp_nn_conv_s8_esp32s3(const int8_t *input,
+                            const uint16_t input_wd,
+                            const uint16_t input_ht,
+                            const uint16_t channels,
+                            const int32_t input_offset,
+                            const uint16_t pad_wd,
+                            const uint16_t pad_ht,
+                            const uint16_t stride_wd,
+                            const uint16_t stride_ht,
+                            const int8_t *filter_data,
+                            const uint16_t filter_wd,
+                            const uint16_t filter_ht,
+                            const int32_t *bias,
+                            int8_t *out_data,
+                            const uint16_t out_wd,
+                            const uint16_t out_ht,
+                            const uint16_t out_channels,
+                            const int32_t out_offset,
+                            const int32_t *out_shift,
+                            const int32_t *out_mult,
+                            const int32_t activation_min,
+                            const int32_t activation_max)
+{
+    int filter_size = filter_wd * filter_ht * channels * out_channels;
+    int input_size = input_wd * input_ht * channels;
+    int align_len = 16 - (filter_size & 15);
+    int16_t *filter_data16 = scratch_buffer;
+    int16_t *input_data16 = scratch_buffer + filter_size + align_len;
+
+    if (scratch_buffer == NULL) {
+        printf("esp_nn_conv error! scratch_buffer not set!\n");
+        return;
+    }
+
+    if (channels % 8 == 0 && filter_wd == 1 && filter_ht == 1 &&
+            pad_wd == 0 && pad_ht == 0 && stride_wd == 1 && stride_ht == 1) {
+        int scratch_offset = (int) (filter_data16 + filter_size);
+        void *scratch_buf = (void *) (scratch_offset + 16 - (scratch_offset & 15));
+        esp_nn_s8_to_s16_esp32s3(filter_data, filter_data16, filter_size);
+        esp_nn_conv_s16_mult8_1x1_esp32s3(
+            input, input_wd, input_ht, channels, input_offset, filter_data16,
+            bias, out_data, out_wd, out_ht, out_channels, out_offset,
+            out_shift, out_mult, activation_min, activation_max, scratch_buf);
+    } else if (channels % 4 == 0 && filter_wd == 1 && filter_ht == 1 &&
+            (input_wd * input_ht) % 16 == 0 && /* TODO: remove this check */
+            pad_wd == 0 && pad_ht == 0 && stride_wd == 1 && stride_ht == 1) {
+        int scratch_offset = (int) (input_data16 + input_size);
+        void *scratch_buf = (void *) (scratch_offset + 16 - (scratch_offset & 15));
+        esp_nn_s8_to_s16_esp32s3(filter_data, filter_data16, filter_size);
+        esp_nn_aligned_s8_to_s16_with_offset_esp32s3(input, input_data16, input_size, input_offset);
+        esp_nn_conv_s16_mult4_1x1_esp32s3(
+            input_data16, input_wd, input_ht, channels, filter_data16,
+            bias, out_data, out_wd, out_ht, out_channels, out_offset,
+            out_shift, out_mult, activation_min, activation_max, scratch_buf);
+    } else if (channels % 8 == 0) {
+        esp_nn_s8_to_s16_esp32s3(filter_data, filter_data16, filter_size);
+        esp_nn_aligned_s8_to_s16_with_offset_esp32s3(input, input_data16, input_size, input_offset);
+        esp_nn_conv_s16_mult8_esp32s3(
+            input_data16, input_wd, input_ht, channels, pad_wd, pad_ht,
+            stride_wd, stride_ht, filter_data16, filter_wd, filter_ht, bias,
+            out_data, out_wd, out_ht, out_channels, out_offset, out_shift,
+            out_mult, activation_min, activation_max);
+    } else if (pad_wd == 0 && pad_ht == 0) {
+        if (filter_wd == 3 && filter_ht == 3 && channels == 3) {
+            esp_nn_conv_s8_pad_valid_ch3_3x3(input, input_wd, input_ht, input_offset,
+                                             stride_wd, stride_ht, filter_data, bias,
+                                             out_data, out_wd, out_ht, out_channels, out_offset,
+                                             out_shift, out_mult, activation_min, activation_max);
+        } else {
+            esp_nn_conv_s8_pad_valid(input, input_wd, input_ht, channels, input_offset,
+                                     stride_wd, stride_ht, filter_data, filter_wd, filter_ht, bias,
+                                     out_data, out_wd, out_ht, out_channels, out_offset, out_shift,
+                                     out_mult, activation_min, activation_max);
+        }
+    } else {
+        /* Basic unrolled version */
+        esp_nn_conv_s8_unrolled(input, input_wd, input_ht, channels, input_offset,
+                                pad_wd, pad_ht, stride_wd, stride_ht,
+                                filter_data, filter_wd, filter_ht, bias,
+                                out_data, out_wd, out_ht, out_channels, out_offset, out_shift,
+                                out_mult, activation_min, activation_max);
+    }
+}
--- a/code/components/esp-nn/src/convolution/esp_nn_depthwise_conv_ansi.c
+++ b/code/components/esp-nn/src/convolution/esp_nn_depthwise_conv_ansi.c
@@ -0,0 +1,97 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+
+#include <common_functions.h>
+
+int esp_nn_get_depthwise_conv_scratch_size_ansi(const uint16_t input_wd,
+                                                const uint16_t input_ht,
+                                                const uint16_t channels,
+                                                const uint16_t ch_mult,
+                                                const uint16_t filter_wd,
+                                                const uint16_t filter_ht)
+{
+    return 0;
+}
+
+void esp_nn_set_depthwise_conv_scratch_buf_ansi(const void *buf)
+{
+
+}
+
+void esp_nn_depthwise_conv_s8_ansi(const int8_t *input_data,
+                                   const uint16_t input_wd,
+                                   const uint16_t input_ht,
+                                   const uint16_t channels,
+                                   const int32_t input_offset,
+                                   const uint16_t pad_wd,
+                                   const uint16_t pad_ht,
+                                   const uint16_t stride_wd,
+                                   const uint16_t stride_ht,
+                                   const uint16_t ch_mult,
+                                   const int8_t *filter_data,
+                                   const uint16_t filter_wd,
+                                   const uint16_t filter_ht,
+                                   const int32_t *bias,
+                                   int8_t *out_data,
+                                   const uint16_t out_wd,
+                                   const uint16_t out_ht,
+                                   const int32_t out_offset,
+                                   const int32_t *out_shift,
+                                   const int32_t *out_mult,
+                                   const int32_t activation_min,
+                                   const int32_t activation_max)
+{
+    int out_idx = 0;
+    for (int out_y = 0; out_y < out_ht; out_y++) { //height loop
+        const int16_t base_y = (out_y * stride_ht) - pad_ht;
+        for (int out_x = 0; out_x < out_wd; out_x++) { //width_loop
+            const int16_t base_x = (out_x * stride_wd) - pad_wd;
+            for (int ch_idx = 0; ch_idx < channels; ch_idx++) {//channel_loop
+                for (int ch_mult_idx = 0; ch_mult_idx < ch_mult; ch_mult_idx++) {
+                    int32_t result = 0;
+                    const int out_ch_idx = ch_mult_idx + ch_idx * ch_mult;
+
+                    /* Select filter so as the point doesn't lie outside block */
+                    int filter_y_start = max(0, -base_y);
+                    int filter_x_start = max(0, -base_x);
+                    int filter_y_end = min(filter_ht, input_ht - base_y);
+                    int filter_x_end = min(filter_wd, input_wd - base_x);
+
+                    for (int filter_y_idx = filter_y_start; filter_y_idx < filter_y_end; filter_y_idx++) {
+                        const int32_t idx_y = base_y + filter_y_idx;
+                        for (int filter_x_idx = filter_x_start; filter_x_idx < filter_x_end; filter_x_idx++) {
+                            const int32_t idx_x = base_x + filter_x_idx;
+                            int32_t input_index = (idx_y * input_wd + idx_x) * channels + ch_idx;
+                            int32_t filter_index = (filter_y_idx * filter_wd + filter_x_idx) * (channels * ch_mult) + out_ch_idx;
+                            int32_t input_val = input_data[input_index] + input_offset;
+                            int32_t filter_val = filter_data[filter_index];
+                            result += input_val * filter_val;
+                        }
+                    }
+                    if (bias) {
+                        result += bias[out_ch_idx];
+                    }
+                    result = esp_nn_multiply_by_quantized_mult(result, out_mult[out_ch_idx], out_shift[out_ch_idx]);
+                    result += out_offset;
+                    result = max(result, activation_min);
+                    result = min(result, activation_max);
+
+                    out_data[out_idx++] = result;
+                }
+            }
+        }
+    }
+}
--- a/code/components/esp-nn/src/convolution/esp_nn_depthwise_conv_s8_esp32s3.c
+++ b/code/components/esp-nn/src/convolution/esp_nn_depthwise_conv_s8_esp32s3.c
@@ -0,0 +1,483 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+#include <stdio.h>
+
+#include <common_functions.h>
+
+static int16_t *scratch_buffer = NULL;
+
+extern void esp_nn_depthwise_conv_s16_mult8_3x3_esp32s3(const int16_t *input_data,
+                                                        const uint16_t input_wd,
+                                                        const uint16_t input_ht,
+                                                        const uint16_t channels,
+                                                        const uint16_t pad_wd,
+                                                        const uint16_t pad_ht,
+                                                        const uint16_t stride_wd,
+                                                        const uint16_t stride_ht,
+                                                        const uint16_t ch_mult,
+                                                        const int16_t *filter_data,
+                                                        const int32_t *bias,
+                                                        int8_t *out_data,
+                                                        const uint16_t out_wd,
+                                                        const uint16_t out_ht,
+                                                        const int32_t out_offset,
+                                                        const int32_t *out_shift,
+                                                        const int32_t *out_mult,
+                                                        const int32_t activation_min,
+                                                        const int32_t activation_max);
+
+extern void esp_nn_depthwise_conv_s8_mult1_3x3_padded_esp32s3(const int8_t *input_data,
+                                                              const uint16_t input_wd,
+                                                              const uint16_t input_ht,
+                                                              const uint16_t channels,
+                                                              const int32_t input_offset,
+                                                              const uint16_t stride_wd,
+                                                              const uint16_t stride_ht,
+                                                              const int8_t *filter_data,
+                                                              const int32_t *bias,
+                                                              int8_t *out_data,
+                                                              const uint16_t out_wd,
+                                                              const uint16_t out_ht,
+                                                              const int32_t out_offset,
+                                                              const int32_t *out_shift,
+                                                              const int32_t *out_mult,
+                                                              const int32_t activation_min,
+                                                              const int32_t activation_max);
+
+extern void esp_nn_depthwise_conv_s16_mult1_3x3_no_pad_esp32s3(const int16_t *input_data,
+                                                               const uint16_t input_wd,
+                                                               const uint16_t input_ht,
+                                                               const uint16_t channels,
+                                                               const uint16_t stride_wd,
+                                                               const uint16_t stride_ht,
+                                                               const int16_t *filter_data,
+                                                               const int32_t *bias,
+                                                               int8_t *out_data,
+                                                               const uint16_t out_wd,
+                                                               const uint16_t out_ht,
+                                                               const int32_t out_offset,
+                                                               const int32_t *out_shift,
+                                                               const int32_t *out_mult,
+                                                               const int32_t activation_min,
+                                                               const int32_t activation_max);
+
+extern void esp_nn_depthwise_conv_s16_mult8_esp32s3(const int16_t *input_data,
+                                                    const uint16_t input_wd,
+                                                    const uint16_t input_ht,
+                                                    const uint16_t channels,
+                                                    const uint16_t pad_wd,
+                                                    const uint16_t pad_ht,
+                                                    const uint16_t stride_wd,
+                                                    const uint16_t stride_ht,
+                                                    const uint16_t ch_mult,
+                                                    const int16_t *filter_data,
+                                                    const uint16_t filter_wd,
+                                                    const uint16_t filter_ht,
+                                                    const int32_t *bias,
+                                                    int8_t *out_data,
+                                                    const uint16_t out_wd,
+                                                    const uint16_t out_ht,
+                                                    const int32_t out_offset,
+                                                    const int32_t *out_shift,
+                                                    const int32_t *out_mult,
+                                                    const int32_t activation_min,
+                                                    const int32_t activation_max);
+
+extern void esp_nn_depthwise_conv_s16_mult4_esp32s3(const int16_t *input_data,
+                                                    const uint16_t input_wd,
+                                                    const uint16_t input_ht,
+                                                    const uint16_t channels,
+                                                    const uint16_t pad_wd,
+                                                    const uint16_t pad_ht,
+                                                    const uint16_t stride_wd,
+                                                    const uint16_t stride_ht,
+                                                    const uint16_t ch_mult,
+                                                    const int16_t *filter_data,
+                                                    const uint16_t filter_wd,
+                                                    const uint16_t filter_ht,
+                                                    const int32_t *bias,
+                                                    int8_t *out_data,
+                                                    const uint16_t out_wd,
+                                                    const uint16_t out_ht,
+                                                    const int32_t out_offset,
+                                                    const int32_t *out_shift,
+                                                    const int32_t *out_mult,
+                                                    const int32_t activation_min,
+                                                    const int32_t activation_max);
+
+extern void esp_nn_depthwise_conv_s16_mult1_3x3_esp32s3(const int16_t *input_data,
+                                                        const uint16_t input_wd,
+                                                        const uint16_t input_ht,
+                                                        const uint16_t channels,
+                                                        const uint16_t pad_wd,
+                                                        const uint16_t pad_ht,
+                                                        const uint16_t stride_wd,
+                                                        const uint16_t stride_ht,
+                                                        const int16_t *filter_data,
+                                                        const int32_t *bias,
+                                                        int8_t *out_data,
+                                                        const uint16_t out_wd,
+                                                        const uint16_t out_ht,
+                                                        const int32_t out_offset,
+                                                        const int32_t *out_shift,
+                                                        const int32_t *out_mult,
+                                                        const int32_t activation_min,
+                                                        const int32_t activation_max);
+
+extern void esp_nn_depthwise_conv_s16_mult1_esp32s3(const int16_t *input_data,
+                                                    const uint16_t input_wd,
+                                                    const uint16_t input_ht,
+                                                    const uint16_t channels,
+                                                    const uint16_t pad_wd,
+                                                    const uint16_t pad_ht,
+                                                    const uint16_t stride_wd,
+                                                    const uint16_t stride_ht,
+                                                    const int16_t *filter_data,
+                                                    const uint16_t filter_wd,
+                                                    const uint16_t filter_ht,
+                                                    const int32_t *bias,
+                                                    int8_t *out_data,
+                                                    const uint16_t out_wd,
+                                                    const uint16_t out_ht,
+                                                    const int32_t out_offset,
+                                                    const int32_t *out_shift,
+                                                    const int32_t *out_mult,
+                                                    const int32_t activation_min,
+                                                    const int32_t activation_max);
+
+extern void esp_nn_s8_to_s16_esp32s3(const int8_t *src, int16_t *dst, const int size);
+
+extern void esp_nn_aligned_s8_to_s16_with_offset_esp32s3(const int8_t *src, int16_t *dst,
+                                                         const int size, const int32_t offset);
+
+static void esp_nn_depthwise_conv_s8_unrolled(const int8_t *input_data,
+                                              const uint16_t input_wd,
+                                              const uint16_t input_ht,
+                                              const uint16_t channels,
+                                              const int32_t input_offset,
+                                              const uint16_t pad_wd,
+                                              const uint16_t pad_ht,
+                                              const uint16_t stride_wd,
+                                              const uint16_t stride_ht,
+                                              const uint16_t ch_mult,
+                                              const int8_t *filter_data,
+                                              const uint16_t filter_wd,
+                                              const uint16_t filter_ht,
+                                              const int32_t *bias,
+                                              int8_t *out_data,
+                                              const uint16_t out_wd,
+                                              const uint16_t out_ht,
+                                              const int32_t out_offset,
+                                              const int32_t *out_shift,
+                                              const int32_t *out_mult,
+                                              const int32_t activation_min,
+                                              const int32_t activation_max)
+{
+    int out_idx = 0;
+    for (int out_y = 0; out_y < out_ht; out_y++) { //height loop
+        const int16_t base_y = (out_y * stride_ht) - pad_ht;
+        for (int out_x = 0; out_x < out_wd; out_x++) { //width_loop
+            const int16_t base_x = (out_x * stride_wd) - pad_wd;
+            for (int ch_idx = 0; ch_idx < channels; ch_idx++) {//channel_loop
+                int ch_mult_idx = 0;
+                for (; ch_mult_idx < ch_mult - 3; ch_mult_idx += 4) {
+                    int32_t result0 = 0, result1 = 0, result2 = 0, result3 = 0;
+                    const int out_ch_idx = ch_mult_idx + ch_idx * ch_mult;
+
+                    /* Select filter so as the point doesn't lie outside block */
+                    int filter_y_start = max(0, -base_y);
+                    int filter_x_start = max(0, -base_x);
+                    int filter_y_end = min(filter_ht, input_ht - base_y);
+                    int filter_x_end = min(filter_wd, input_wd - base_x);
+
+                    for (int filter_y_idx = filter_y_start; filter_y_idx < filter_y_end; filter_y_idx++) {
+                        const int32_t idx_y = base_y + filter_y_idx;
+                        for (int filter_x_idx = filter_x_start; filter_x_idx < filter_x_end; filter_x_idx++) {
+                            const int32_t idx_x = base_x + filter_x_idx;
+                            int32_t input_index = (idx_y * input_wd + idx_x) * channels + ch_idx;
+                            int32_t filter_index = (filter_y_idx * filter_wd + filter_x_idx) * (channels * ch_mult) + out_ch_idx;
+                            int32_t input_val = input_data[input_index] + input_offset;
+                            int32_t filter_val0 = filter_data[filter_index + 0];
+                            int32_t filter_val1 = filter_data[filter_index + 1];
+                            int32_t filter_val2 = filter_data[filter_index + 2];
+                            int32_t filter_val3 = filter_data[filter_index + 3];
+                            result0 += input_val * filter_val0;
+                            result1 += input_val * filter_val1;
+                            result2 += input_val * filter_val2;
+                            result3 += input_val * filter_val3;
+                        }
+                    }
+                    if (bias) {
+                        result0 += bias[out_ch_idx + 0];
+                        result1 += bias[out_ch_idx + 1];
+                        result2 += bias[out_ch_idx + 2];
+                        result3 += bias[out_ch_idx + 3];
+                    }
+                    result0 = esp_nn_multiply_by_quantized_mult(result0,
+                                out_mult[out_ch_idx + 0], out_shift[out_ch_idx + 0]);
+                    result1 = esp_nn_multiply_by_quantized_mult(result1,
+                                out_mult[out_ch_idx + 1], out_shift[out_ch_idx + 1]);
+                    result2 = esp_nn_multiply_by_quantized_mult(result2,
+                                out_mult[out_ch_idx + 2], out_shift[out_ch_idx + 2]);
+                    result3 = esp_nn_multiply_by_quantized_mult(result3,
+                                out_mult[out_ch_idx + 3], out_shift[out_ch_idx + 3]);
+
+                    result0 += out_offset;
+                    result1 += out_offset;
+                    result2 += out_offset;
+                    result3 += out_offset;
+
+                    result0 = max(result0, activation_min);
+                    result1 = max(result1, activation_min);
+                    result2 = max(result2, activation_min);
+                    result3 = max(result3, activation_min);
+
+                    result0 = min(result0, activation_max);
+                    result1 = min(result1, activation_max);
+                    result2 = min(result2, activation_max);
+                    result3 = min(result3, activation_max);
+
+                    out_data[out_idx++] = result0;
+                    out_data[out_idx++] = result1;
+                    out_data[out_idx++] = result2;
+                    out_data[out_idx++] = result3;
+                }
+
+                /* left-over */
+                for (; ch_mult_idx < ch_mult; ch_mult_idx++) {
+                    int32_t result = 0;
+                    const int out_ch_idx = ch_mult_idx + ch_idx * ch_mult;
+
+                    /* Select filter so as the point doesn't lie outside block */
+                    int filter_y_start = max(0, -base_y);
+                    int filter_x_start = max(0, -base_x);
+                    int filter_y_end = min(filter_ht, input_ht - base_y);
+                    int filter_x_end = min(filter_wd, input_wd - base_x);
+
+                    for (int filter_y_idx = filter_y_start; filter_y_idx < filter_y_end; filter_y_idx++) {
+                        const int32_t idx_y = base_y + filter_y_idx;
+                        for (int filter_x_idx = filter_x_start; filter_x_idx < filter_x_end; filter_x_idx++) {
+                            const int32_t idx_x = base_x + filter_x_idx;
+                            int32_t input_index = (idx_y * input_wd + idx_x) * channels + ch_idx;
+                            int32_t filter_index = (filter_y_idx * filter_wd + filter_x_idx) * (channels * ch_mult) + out_ch_idx;
+                            int32_t input_val = input_data[input_index] + input_offset;
+                            int32_t filter_val = filter_data[filter_index];
+                            result += input_val * filter_val;
+                        }
+                    }
+                    if (bias) {
+                        result += bias[out_ch_idx];
+                    }
+                    result = esp_nn_multiply_by_quantized_mult(result, out_mult[out_ch_idx], out_shift[out_ch_idx]);
+                    result += out_offset;
+                    result = max(result, activation_min);
+                    result = min(result, activation_max);
+
+                    out_data[out_idx++] = result;
+                }
+            }
+        }
+    }
+}
+
+void esp_nn_depthwise_conv_s8_ch_mult1(const int8_t *input_data,
+                                       const uint16_t input_wd,
+                                       const uint16_t input_ht,
+                                       const uint16_t channels,
+                                       const int32_t input_offset,
+                                       const uint16_t pad_wd,
+                                       const uint16_t pad_ht,
+                                       const uint16_t stride_wd,
+                                       const uint16_t stride_ht,
+                                       const int8_t *filter_data,
+                                       const uint16_t filter_wd,
+                                       const uint16_t filter_ht,
+                                       const int32_t *bias,
+                                       int8_t *out_data,
+                                       const uint16_t out_wd,
+                                       const uint16_t out_ht,
+                                       const int32_t out_offset,
+                                       const int32_t *out_shift,
+                                       const int32_t *out_mult,
+                                       const int32_t activation_min,
+                                       const int32_t activation_max)
+{
+    int out_idx = 0;
+    for (int out_y = 0; out_y < out_ht; out_y++) { //height loop
+        const int16_t base_y = (out_y * stride_ht) - pad_ht;
+        for (int out_x = 0; out_x < out_wd; out_x++) { //width_loop
+            const int16_t base_x = (out_x * stride_wd) - pad_wd;
+            for (int ch_idx = 0; ch_idx < channels; ch_idx++) {//channel_loop
+                int32_t result = 0;
+                /* Select filter so as the point doesn't lie outside block */
+                int filter_y_start = max(0, -base_y);
+                int filter_x_start = max(0, -base_x);
+                int filter_y_end = min(filter_ht, input_ht - base_y);
+                int filter_x_end = min(filter_wd, input_wd - base_x);
+
+                for (int filter_y_idx = filter_y_start; filter_y_idx < filter_y_end; filter_y_idx++) {
+                    const int32_t idx_y = base_y + filter_y_idx;
+                    for (int filter_x_idx = filter_x_start; filter_x_idx < filter_x_end; filter_x_idx++) {
+                        const int32_t idx_x = base_x + filter_x_idx;
+                        int32_t input_index = (idx_y * input_wd + idx_x) * channels + ch_idx;
+                        int32_t filter_index = (filter_y_idx * filter_wd + filter_x_idx) * channels + ch_idx;
+                        int32_t input_val = input_data[input_index] + input_offset;
+                        int32_t filter_val = filter_data[filter_index];
+                        result += input_val * filter_val;
+                    }
+                }
+                if (bias) {
+                    result += bias[ch_idx];
+                }
+                result = esp_nn_multiply_by_quantized_mult(result, out_mult[ch_idx], out_shift[ch_idx]);
+                result += out_offset;
+                result = max(result, activation_min);
+                result = min(result, activation_max);
+
+                out_data[out_idx++] = result;
+            }
+        }
+    }
+}
+
+int esp_nn_get_depthwise_conv_scratch_size_esp32s3(const uint16_t input_wd,
+                                                   const uint16_t input_ht,
+                                                   const uint16_t channels,
+                                                   const uint16_t ch_mult,
+                                                   const uint16_t filter_wd,
+                                                   const uint16_t filter_ht)
+{
+    int filter_size = filter_wd * filter_ht * channels * ch_mult;
+    int padding_used = ((filter_wd == 3) && (filter_ht == 3)) * 2;
+    int input_size = (input_wd + padding_used) * (input_ht + padding_used) * channels;
+    return  2 * (filter_size + input_size) + 16; //16 for alignment
+}
+
+void esp_nn_set_depthwise_conv_scratch_buf_esp32s3(void *buf)
+{
+    scratch_buffer = (int16_t *) buf;
+}
+
+/**
+ * Assumption 1: i/p channels == o/p channels
+ * Assumption 2: Pointers are valid
+ * Assumption 3: dialation width = 1
+ */
+void esp_nn_depthwise_conv_s8_esp32s3(const int8_t *input_data,
+                                      const uint16_t input_wd,
+                                      const uint16_t input_ht,
+                                      const uint16_t channels,
+                                      const int32_t input_offset,
+                                      const uint16_t pad_wd,
+                                      const uint16_t pad_ht,
+                                      const uint16_t stride_wd,
+                                      const uint16_t stride_ht,
+                                      const uint16_t ch_mult,
+                                      const int8_t *filter_data,
+                                      const uint16_t filter_wd,
+                                      const uint16_t filter_ht,
+                                      const int32_t *bias,
+                                      int8_t *out_data,
+                                      const uint16_t out_wd,
+                                      const uint16_t out_ht,
+                                      const int32_t out_offset,
+                                      const int32_t *out_shift,
+                                      const int32_t *out_mult,
+                                      const int32_t activation_min,
+                                      const int32_t activation_max)
+{
+    int filter_size = filter_wd * filter_ht * channels * ch_mult;
+    int align_len = 16 - (filter_size & 15);
+    int input_size = input_wd * input_ht * channels;
+    int16_t *filter_data16 = scratch_buffer;
+    int16_t *input_data16 = scratch_buffer + filter_size + align_len;
+    if (scratch_buffer == NULL) {
+        printf("esp_nn_depthwise_conv error! scratch_buffer not set!\n");
+        return;
+    }
+
+    if ((ch_mult == 1) && (channels % 8 == 0)) {
+        if ((filter_wd == 3) && (filter_ht == 3)) {
+            if ((channels % 16 == 0) && (pad_wd == 1) && (pad_ht == 1)) {
+                /* process in 8 bits */
+                int8_t *filter_aligned = (int8_t *) scratch_buffer;
+                int8_t *input_padded = (int8_t *) scratch_buffer + filter_size + align_len;
+                memcpy(filter_aligned, filter_data, filter_size);
+                esp_nn_aligned_s8_pad_with_value(input_data, input_padded, input_wd, input_ht, channels,
+                                                 -input_offset, pad_wd, pad_ht);
+                esp_nn_depthwise_conv_s8_mult1_3x3_padded_esp32s3(input_padded, input_wd + 2 * pad_wd,
+                                                                  input_ht + 2 * pad_ht, channels, input_offset,
+                                                                  stride_wd, stride_ht, filter_aligned, bias,
+                                                                  out_data, out_wd, out_ht, out_offset, out_shift,
+                                                                  out_mult, activation_min, activation_max);
+            } else if ((pad_wd == 0) && (pad_ht == 0) &&
+                    // because this does not handle padding offset cases yet, run just for stride (1, 1).
+                    // end padding of input with `-input_offset` should solve this
+                    (stride_wd == 1) && (stride_ht == 1)) {
+                /* process in 8 bits */
+                int8_t *filter_aligned = (int8_t *) scratch_buffer;
+                memcpy(filter_aligned, filter_data, filter_size);
+                esp_nn_depthwise_conv_s8_mult1_3x3_padded_esp32s3(input_data, input_wd, input_ht, channels, input_offset,
+                                                                  stride_wd, stride_ht, filter_aligned,
+                                                                  bias, out_data, out_wd, out_ht, out_offset, out_shift,
+                                                                  out_mult, activation_min, activation_max);
+            } else { /* (channels % 8) == 0 && pad_wd == 1 && pad_ht == 1 */
+                esp_nn_s8_to_s16_esp32s3(filter_data, filter_data16, filter_size);
+                esp_nn_aligned_s8_to_s16_with_offset_esp32s3(input_data, input_data16, input_size, input_offset);
+                esp_nn_depthwise_conv_s16_mult1_3x3_esp32s3(input_data16, input_wd, input_ht, channels,
+                                                            pad_wd, pad_ht, stride_wd, stride_ht, filter_data16,
+                                                            bias, out_data, out_wd, out_ht, out_offset, out_shift,
+                                                            out_mult, activation_min, activation_max);
+            }
+        } else { // all other ch_mult == 1, `channels % 8 == 0`
+            esp_nn_depthwise_conv_s8_ch_mult1(input_data, input_wd, input_ht, channels, input_offset,
+                                              pad_wd, pad_ht, stride_wd, stride_ht,
+                                              filter_data, filter_wd, filter_ht,
+                                              bias, out_data, out_wd, out_ht, out_offset, out_shift,
+                                              out_mult, activation_min, activation_max);
+        }
+    } else if (ch_mult % 8 == 0) {
+        esp_nn_s8_to_s16_esp32s3(filter_data, filter_data16, filter_size);
+        esp_nn_aligned_s8_to_s16_with_offset_esp32s3(input_data, input_data16, input_size, input_offset);
+        if (filter_wd == 3 && filter_ht == 3) {
+            esp_nn_depthwise_conv_s16_mult8_3x3_esp32s3(input_data16, input_wd, input_ht, channels,
+                                                        pad_wd, pad_ht, stride_wd, stride_ht, ch_mult,
+                                                        filter_data16, bias,
+                                                        out_data, out_wd, out_ht, out_offset, out_shift,
+                                                        out_mult, activation_min, activation_max);
+        } else {
+            esp_nn_depthwise_conv_s16_mult8_esp32s3(input_data16, input_wd, input_ht, channels,
+                                                    pad_wd, pad_ht, stride_wd, stride_ht, ch_mult,
+                                                    filter_data16, filter_wd, filter_ht, bias,
+                                                    out_data, out_wd, out_ht, out_offset, out_shift,
+                                                    out_mult, activation_min, activation_max);
+        }
+    } else if (ch_mult % 4 == 0) {
+        esp_nn_s8_to_s16_esp32s3(filter_data, filter_data16, filter_size);
+        esp_nn_aligned_s8_to_s16_with_offset_esp32s3(input_data, input_data16, input_size, input_offset);
+        esp_nn_depthwise_conv_s16_mult4_esp32s3(input_data16, input_wd, input_ht, channels,
+                                                pad_wd, pad_ht, stride_wd, stride_ht, ch_mult,
+                                                filter_data16, filter_wd, filter_ht, bias,
+                                                out_data, out_wd, out_ht, out_offset, out_shift,
+                                                out_mult, activation_min, activation_max);
+    } else {
+        esp_nn_depthwise_conv_s8_unrolled(input_data, input_wd, input_ht, channels, input_offset,
+                                          pad_wd, pad_ht, stride_wd, stride_ht, ch_mult,
+                                          filter_data, filter_wd, filter_ht,
+                                          bias, out_data, out_wd, out_ht, out_offset, out_shift,
+                                          out_mult, activation_min, activation_max);
+    }
+}
--- a/code/components/esp-nn/src/fully_connected/esp_nn_fully_connected_ansi.c
+++ b/code/components/esp-nn/src/fully_connected/esp_nn_fully_connected_ansi.c
@@ -0,0 +1,50 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+
+#include <common_functions.h>
+
+void esp_nn_fully_connected_s8_ansi(const int8_t *input_data,
+                                    const int32_t input_offset,
+                                    const uint16_t row_len,
+                                    const int8_t *filter_data,
+                                    const int32_t filter_offset,
+                                    const int32_t *bias,
+                                    int8_t *out_data,
+                                    const uint16_t out_channels,
+                                    const int32_t out_offset,
+                                    const int32_t out_shift,
+                                    const int32_t out_mult,
+                                    const int32_t activation_min,
+                                    const int32_t activation_max)
+{
+    for (int32_t out_c = 0; out_c < out_channels; ++out_c) {
+        int32_t result = 0;
+        for (int32_t data_idx = 0; data_idx < row_len; data_idx++) {
+            int32_t filter_index = row_len * out_c + data_idx;
+            int32_t input_val = input_data[data_idx];
+            int32_t filter_val = filter_data[filter_index];
+            result += (filter_val + filter_offset) * (input_val + input_offset);
+        }
+        if (bias) {
+            result += bias[out_c];
+        }
+        result = esp_nn_multiply_by_quantized_mult(result, out_mult, out_shift);
+        result += out_offset;
+        result = max(result, activation_min);
+        result = min(result, activation_max);
+        out_data[out_c] = (int8_t) result;
+    }
+}
--- a/code/components/esp-nn/src/pooling/esp_nn_avg_pool_ansi.c
+++ b/code/components/esp-nn/src/pooling/esp_nn_avg_pool_ansi.c
@@ -0,0 +1,72 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+
+#include <common_functions.h>
+
+void esp_nn_avg_pool_s8_ansi(const int8_t *input,
+                             const uint16_t input_wd,
+                             const uint16_t input_ht,
+                             int8_t *output,
+                             const uint16_t output_wd,
+                             const uint16_t output_ht,
+                             const uint16_t stride_wd,
+                             const uint16_t stride_ht,
+                             const uint16_t filter_wd,
+                             const uint16_t filter_ht,
+                             const uint16_t pad_wd,
+                             const uint16_t pad_ht,
+                             const int32_t activation_min,
+                             const int32_t activation_max,
+                             const uint16_t channels)
+{
+    int32_t base_y = -pad_ht;
+    for (int32_t out_y = 0; out_y < output_ht; out_y++, base_y += stride_ht) {
+        int32_t base_x = -pad_wd;
+        for (int32_t out_x = 0; out_x < output_wd; out_x++, base_x += stride_wd) {
+            for (int32_t ch_idx = 0; ch_idx < channels; ch_idx++) {
+                int32_t result = 0;
+                int32_t filter_cnt = 0;
+                /* Make sure filter does not cross the input box */
+                int32_t filter_y_start = max(0, -base_y);
+                int32_t filter_x_start = max(0, -base_x);
+
+                int32_t filter_y_end = min(filter_ht, input_ht - base_y);
+                int32_t filter_x_end = min(filter_wd, input_wd - base_x);
+
+                for (int32_t filter_y = filter_y_start; filter_y < filter_y_end; filter_y++) {
+                    for (int32_t filter_x = filter_x_start; filter_x < filter_x_end; filter_x++) {
+                        int32_t in_x_idx = base_x + filter_x;
+                        int32_t in_y_idx = base_y + filter_y;
+                        int32_t input_index = (in_y_idx * input_wd + in_x_idx) * channels + ch_idx;
+                        result += input[input_index];
+                        filter_cnt++;
+                    }
+                }
+
+                /* Rounded average */
+                result = result > 0 ? (result + filter_cnt / 2) / filter_cnt
+                                    : (result - filter_cnt / 2) / filter_cnt;
+
+                /* Activation function */
+                result = max(result, activation_min);
+                result = min(result, activation_max);
+
+                int32_t output_index = (out_y * output_wd + out_x) * channels + ch_idx;
+                output[output_index] = (int8_t) result;
+            }
+        }
+    }
+}
--- a/code/components/esp-nn/src/pooling/esp_nn_max_pool_ansi.c
+++ b/code/components/esp-nn/src/pooling/esp_nn_max_pool_ansi.c
@@ -0,0 +1,66 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+
+#include <common_functions.h>
+
+void esp_nn_max_pool_s8_ansi(const int8_t *input,
+                             const uint16_t input_wd,
+                             const uint16_t input_ht,
+                             int8_t *output,
+                             const uint16_t output_wd,
+                             const uint16_t output_ht,
+                             const uint16_t stride_wd,
+                             const uint16_t stride_ht,
+                             const uint16_t filter_wd,
+                             const uint16_t filter_ht,
+                             const uint16_t pad_wd,
+                             const uint16_t pad_ht,
+                             const int32_t activation_min,
+                             const int32_t activation_max,
+                             const uint16_t channels)
+{
+    int32_t base_y = -pad_ht;
+    for (int32_t out_y = 0; out_y < output_ht; out_y++, base_y += stride_ht) {
+        int32_t base_x = -pad_wd;
+        for (int32_t out_x = 0; out_x < output_wd; out_x++, base_x += stride_wd) {
+            /* Make sure filter does not cross the input box */
+            int32_t filter_y_start = max(0, -base_y);
+            int32_t filter_x_start = max(0, -base_x);
+            int32_t filter_y_end = min(filter_ht, input_ht - base_y);
+            int32_t filter_x_end = min(filter_wd, input_wd - base_x);
+
+            for (int32_t ch_idx = 0; ch_idx < channels; ch_idx++) {
+                int8_t result = INT8_MIN;
+
+                for (int32_t filter_y = filter_y_start; filter_y < filter_y_end; filter_y++) {
+                    for (int32_t filter_x = filter_x_start; filter_x < filter_x_end; filter_x++) {
+                        int32_t in_x_idx = base_x + filter_x;
+                        int32_t in_y_idx = base_y + filter_y;
+                        int32_t input_index = (in_y_idx * input_wd + in_x_idx) * channels + ch_idx;
+                        result = max(input[input_index], result);
+                    }
+                }
+
+                /* Activation function */
+                result = max(result, activation_min);
+                result = min(result, activation_max);
+
+                int32_t output_index = (out_y * output_wd + out_x) * channels + ch_idx;
+                output[output_index] = result;
+            }
+        }
+    }
+}
--- a/code/components/esp-nn/src/softmax/esp_nn_softmax_ansi.c
+++ b/code/components/esp-nn/src/softmax/esp_nn_softmax_ansi.c
@@ -0,0 +1,88 @@
+// Copyright 2022 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "softmax_common.h"
+
+int32_t esp_nn_get_softmax_scratch_size_ansi(const int32_t width, const int32_t height)
+{
+    (void) width;
+    (void) height;
+    return 0;
+}
+
+void esp_nn_set_softmax_scratch_buf_ansi(void *buffer)
+{
+    (void) buffer;
+    return;
+}
+
+void esp_nn_softmax_s8_ansi(const int8_t *input_data,
+                            const int32_t height,
+                            const int32_t width,
+                            const int32_t mult,
+                            const int32_t shift,
+                            const int32_t diff_min,
+                            int8_t *output_data)
+{
+    // The representation chosen for the input to the exp() function is Q5.26.
+    // We need to leave extra space since values that we skip might be as large as
+    // -32 before multiplying by input mult, and therefore as large as
+    // -16 afterwards.  Note that exp(-8) is definitely not insignificant to
+    // accumulation, but exp(-16) definitely is.
+#define ACCUM_BITS  12
+#define DIFF_BITS   5
+
+    const int32_t mask = (1 << shift);
+    int32_t col = 0;
+    const int8_t *in_ptr = input_data;
+    int8_t *out_ptr = output_data;
+
+    for (int row_idx = 0; row_idx < height; row_idx++) {
+        int8_t max_in_row = in_ptr[0];
+        for (col = 1; col < width; col++) {
+            max_in_row = max(max_in_row, in_ptr[col]);
+        }
+
+        int32_t input_diff = 0;
+        int32_t sum_of_exps = 0;
+
+        for (col = 0; col < width; col++) {
+            input_diff = in_ptr[col] - max_in_row;
+            if (input_diff >= diff_min) {
+                const int32_t input_diff_rescaled = SAT_HIGH_MUL(input_diff * mask, mult);
+                const int32_t exp_raw = esp_nn_exp_on_negative_values(input_diff_rescaled);
+                sum_of_exps += DIV_POW2(exp_raw, ACCUM_BITS);
+            }
+        }
+
+        const int32_t headroom_plus1 = esp_nn_clz32((uint32_t) sum_of_exps);
+        const int32_t shifted_scale = ONE_OVER_ONE_X((sum_of_exps << headroom_plus1) - (1 << 31));
+        const int32_t bits_over_unit = ACCUM_BITS - headroom_plus1 + 31 - sizeof(int8_t) * 8;
+
+        for (col = 0; col < width; col++) {
+            input_diff = in_ptr[col] - max_in_row;
+            if (input_diff >= diff_min) {
+                const int32_t input_diff_rescaled = SAT_HIGH_MUL(input_diff * mask, mult);
+                const int32_t exp_raw = esp_nn_exp_on_negative_values(input_diff_rescaled);
+                const int32_t shifted_output = SAT_HIGH_MUL(shifted_scale, exp_raw);
+                const int32_t result = DIV_POW2(shifted_output, bits_over_unit) - 128;
+                out_ptr[col] = (int8_t) esp_nn_saturate8(result);
+            } else {
+                out_ptr[col] = -128;
+            }
+        }
+        in_ptr  += width;
+        out_ptr += width;
+    }
+}
--- a/code/components/esp-nn/src/softmax/esp_nn_softmax_opt.c
+++ b/code/components/esp-nn/src/softmax/esp_nn_softmax_opt.c
@@ -0,0 +1,108 @@
+// Copyright 2022 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include "softmax_common.h"
+#include <stdio.h>
+
+static int32_t *scratch_buf = NULL;
+
+/**
+ * @brief   Get scratch buffer size needed by softmax function
+ *
+ * @param   width
+ * @param   height
+ * @return  size in bytes
+ *
+ * @note    buffer must be 4 byte aligned
+ */
+int32_t esp_nn_get_softmax_scratch_size_opt(const int32_t width, const int32_t height)
+{
+    (void) height;
+    return width * 4;
+}
+
+/**
+ * @brief   Set scratch buffer to be used by softmax function
+ *
+ * @param   buffer  this can be NULL if one needs to unset it
+ *                  must be aligned to 4 bytes
+ */
+void esp_nn_set_softmax_scratch_buf_opt(void *buffer)
+{
+    scratch_buf = (int32_t *) buffer;
+}
+
+void esp_nn_softmax_s8_opt(const int8_t *input_data,
+                           const int32_t height,
+                           const int32_t width,
+                           const int32_t mult,
+                           const int32_t shift,
+                           const int32_t diff_min,
+                           int8_t *output_data)
+{
+    if (scratch_buf == NULL) {
+        printf("%s error! scratch buffer not set\n", __FUNCTION__);
+        return;
+    }
+    // The representation chosen for the input to the exp() function is Q5.26.
+    // We need to leave extra space since values that we skip might be as large as
+    // -32 before multiplying by input mult, and therefore as large as
+    // -16 afterwards.  Note that exp(-8) is definitely not insignificant to
+    // accumulation, but exp(-16) definitely is.
+#define ACCUM_BITS  12
+#define DIFF_BITS   5
+
+    const int32_t mask = (1 << shift);
+    int32_t col = 0;
+    const int8_t *in_ptr = input_data;
+    int8_t *out_ptr = output_data;
+
+    for (int row_idx = 0; row_idx < height; row_idx++) {
+        int8_t max_in_row = in_ptr[0];
+        for (col = 1; col < width; col++) {
+            max_in_row = max(max_in_row, in_ptr[col]);
+        }
+
+        int32_t input_diff = 0;
+        int32_t sum_of_exps = 0;
+
+        for (col = 0; col < width; col++) {
+            input_diff = in_ptr[col] - max_in_row;
+            if (input_diff >= diff_min) {
+                const int32_t input_diff_rescaled = SAT_HIGH_MUL(input_diff * mask, mult);
+                const int32_t exp_raw = esp_nn_exp_on_negative_values(input_diff_rescaled);
+                scratch_buf[col] = exp_raw; // store to avoid duplicate calculation later
+                sum_of_exps += DIV_POW2(exp_raw, ACCUM_BITS);
+            }
+        }
+
+        const int32_t headroom_plus1 = esp_nn_clz32((uint32_t) sum_of_exps);
+        const int32_t shifted_scale = ONE_OVER_ONE_X((sum_of_exps << headroom_plus1) - (1 << 31));
+        const int32_t bits_over_unit = ACCUM_BITS - headroom_plus1 + 31 - sizeof(int8_t) * 8;
+
+        for (col = 0; col < width; col++) {
+            input_diff = in_ptr[col] - max_in_row;
+            if (input_diff >= diff_min) {
+                int32_t exp_raw = scratch_buf[col];
+                const int32_t shifted_output = SAT_HIGH_MUL(shifted_scale, exp_raw);
+                const int32_t result = DIV_POW2(shifted_output, bits_over_unit) - 128;
+                out_ptr[col] = (int8_t) esp_nn_saturate8(result);
+            } else {
+                out_ptr[col] = -128;
+            }
+        }
+        in_ptr  += width;
+        out_ptr += width;
+    }
+}
--- a/code/components/esp-nn/src/softmax/softmax_common.h
+++ b/code/components/esp-nn/src/softmax/softmax_common.h
@@ -0,0 +1,104 @@
+// Copyright 2022 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+#include <common_functions.h>
+
+#define MASK_IF_ZERO(x)                 (x) == 0 ? ~0 : 0
+#define MASK_IF_NON_ZERO(x)             (x) != 0 ? ~0 : 0
+#define SELECT_USING_MASK(mask, a, b)   ((mask) & (a)) ^ (~(mask) & (b))
+#define SAT_HIGH_MUL(x, y)              esp_nn_sat_round_doubling_high_mul((x), (y))
+#define DIV_POW2(x,y)                   esp_nn_div_by_power_of_two((x), (y))
+
+__NN_FORCE_INLINE__ int32_t mul_power_of_2(int val, int exp)
+{
+    const int32_t thresh = ((1 << (31 - exp)) - 1);
+    int32_t result = val << exp;
+    result = SELECT_USING_MASK(MASK_IF_NON_ZERO(val > thresh), INT32_MAX, result);
+    result = SELECT_USING_MASK(MASK_IF_NON_ZERO(val < -thresh), INT32_MIN, result);
+    return result;
+}
+
+/**
+ * @brief   Calculate `1 / (1 + x)` for x in [0, 1]
+ *
+ * @param   val     input value to calculate `1/(1+x)` for
+ * @return  `int32_t` result
+ * @note    Newton-Raphson division
+ *
+ *          https://en.wikipedia.org/wiki/Division_algorithm#Newton.E2.80.93Raphson_division
+ *          Refer to that page for the logic behind the 48/17 and 32/17 constants.
+ *          Pseudocode: https://en.wikipedia.org/wiki/Division_algorithm#Pseudocode
+ */
+__NN_FORCE_INLINE__ int32_t esp_nn_one_over_one_plus_x_for_x_in_0_1(int32_t val)
+{
+    const int64_t sum = (int64_t) val + INT32_MAX;
+    const int32_t half_denominator = (int32_t) ((sum + (sum >= 0 ? 1 : -1)) / 2L);
+    int32_t constant_48_over_17 = 1515870810;
+    int32_t constant_neg_32_over_17 = -1010580540;
+    int32_t x = constant_48_over_17 + SAT_HIGH_MUL(half_denominator, constant_neg_32_over_17);
+    const int32_t fixed_2_one = (1 << 29);
+
+    x += mul_power_of_2(SAT_HIGH_MUL(x, fixed_2_one - SAT_HIGH_MUL(half_denominator, x)), 2);
+    x += mul_power_of_2(SAT_HIGH_MUL(x, fixed_2_one - SAT_HIGH_MUL(half_denominator, x)), 2);
+    x += mul_power_of_2(SAT_HIGH_MUL(x, fixed_2_one - SAT_HIGH_MUL(half_denominator, x)), 2);
+
+    return mul_power_of_2(x, 1);
+}
+
+#define ONE_OVER_ONE_X(x)   esp_nn_one_over_one_plus_x_for_x_in_0_1((x))
+
+/**
+ * @brief   Return exp(x) for x < 0.
+ *
+ */
+__NN_FORCE_INLINE__ int32_t esp_nn_exp_on_negative_values(int32_t val)
+{
+    int32_t shift = 24;
+
+    const int32_t one_quarter = (1 << shift);
+    int32_t mask = one_quarter - 1;
+    const int32_t val_mod_minus_quarter = (val & mask) - one_quarter;
+    const int32_t remainder             = val_mod_minus_quarter - val;
+
+    // calculate exponent for x in [-1/4, 0) in `result`
+    const int32_t x                     = (val_mod_minus_quarter << 5) + (1 << 28);
+    const int32_t x2                    = SAT_HIGH_MUL(x, x);
+    const int32_t x3                    = SAT_HIGH_MUL(x2, x);
+    const int32_t x4                    = SAT_HIGH_MUL(x2, x2);
+    const int32_t one_over_3            = 715827883;
+    const int32_t one_over_8            = 1895147668;
+
+    const int32_t x4_over_4 = DIV_POW2(x4, 2);
+    const int32_t x4_over_4_plus_x3_over_6_plus_x2_over_2 = DIV_POW2(SAT_HIGH_MUL(x4_over_4 + x3, one_over_3) + x2, 1);
+    int32_t result = one_over_8 + SAT_HIGH_MUL(one_over_8, x + x4_over_4_plus_x3_over_6_plus_x2_over_2);
+
+#define SELECT_IF_NON_ZERO(x) {                                   \
+    mask   = MASK_IF_NON_ZERO(remainder & (1 << shift++));        \
+    result = SELECT_USING_MASK(mask, SAT_HIGH_MUL(result, x), result); \
+}
+
+    SELECT_IF_NON_ZERO(1672461947)
+    SELECT_IF_NON_ZERO(1302514674)
+    SELECT_IF_NON_ZERO(790015084)
+    SELECT_IF_NON_ZERO(290630308)
+    SELECT_IF_NON_ZERO(39332535)
+    SELECT_IF_NON_ZERO(720401)
+    SELECT_IF_NON_ZERO(242)
+
+#undef SELECT_IF_NON_ZERO
+
+    mask = MASK_IF_ZERO(val);
+    return SELECT_USING_MASK(mask, INT32_MAX, result);
+}
--- a/code/components/esp-nn/test_app/CMakeLists.txt
+++ b/code/components/esp-nn/test_app/CMakeLists.txt
@@ -0,0 +1,9 @@
+# The following lines of boilerplate have to be in your project's
+# CMakeLists in this exact order for cmake to work correctly
+cmake_minimum_required(VERSION 3.5)
+
+set(EXTRA_COMPONENT_DIRS "../" "../tests/")
+set(IDF_EXCLUDE_COMPONENTS test test_app)
+
+include($ENV{IDF_PATH}/tools/cmake/project.cmake)
+project(test_app)
--- a/code/components/esp-nn/test_app/main/CMakeLists.txt
+++ b/code/components/esp-nn/test_app/main/CMakeLists.txt
@@ -0,0 +1,7 @@
+
+set(COMPONENT_SRCS "main.c")
+set(COMPONENT_ADD_INCLUDEDIRS "")
+
+set(COMPONENT_PRIV_REQUIRES tests)
+
+register_component()
--- a/code/components/esp-nn/test_app/main/component.mk
+++ b/code/components/esp-nn/test_app/main/component.mk
@@ -0,0 +1,8 @@
+#
+# Main component makefile.
+#
+# This Makefile can be left empty. By default, it will take the sources in the 
+# src/ directory, compile them and link them into lib(subdirectory_name).a
+# in the build directory. This behaviour is entirely configurable,
+# please read the ESP-IDF documents if you need to do this.
+# 
--- a/code/components/esp-nn/test_app/main/main.c
+++ b/code/components/esp-nn/test_app/main/main.c
@@ -0,0 +1,87 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <freertos/FreeRTOS.h>
+#include <freertos/task.h>
+#include <esp_log.h>
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+
+#include <test_functions.h>
+#include <esp_timer.h>
+
+static const char *TAG = "test_app";
+static uint32_t start_c, start_opt, total_c, total_opt;
+
+void profile_c_start()
+{
+    /* initiate profiling */
+    start_c = esp_cpu_get_ccount();
+}
+
+void profile_c_end()
+{
+    /* record profile number */
+    total_c = esp_cpu_get_ccount() - start_c;
+}
+
+void profile_opt_start()
+{
+    /* initiate profiling */
+    start_opt = esp_cpu_get_ccount();
+}
+
+void profile_opt_end()
+{
+    /* record profile number */
+    total_opt = esp_cpu_get_ccount() - start_opt;
+}
+
+void app_main()
+{
+    /* s8 tests */
+    ESP_LOGI(TAG, "Running s8 tests...");
+    esp_nn_add_elementwise_s8_test();
+    printf("add, c %u opt %u\n", total_c, total_opt);
+    esp_nn_mul_elementwise_s8_test();
+    printf("mul, c %u opt %u\n", total_c, total_opt);
+    esp_nn_depthwise_conv_s8_test();
+    printf("depthwise, c %u opt %u\n", total_c, total_opt);
+    esp_nn_conv_s8_test();
+    printf("conv2d, c %u opt %u\n", total_c, total_opt);
+
+    esp_nn_relu6_s8_test();
+    printf("relu, c %u opt %u\n", total_c, total_opt);
+    esp_nn_avg_pool_s8_test();
+    printf("avg_pool, c %u opt %u\n", total_c, total_opt);
+    esp_nn_max_pool_s8_test();
+    printf("max_pool, c %u opt %u\n", total_c, total_opt);
+    esp_nn_fully_connected_s8_test();
+    printf("fully_connected, c %u opt %u\n", total_c, total_opt);
+    esp_nn_softmax_s8_test();
+    printf("softmax, c %u opt %u\n", total_c, total_opt);
+    ESP_LOGI(TAG, "s8 tests done!\n");
+
+    /* u8 tests */
+    //ESP_LOGI(TAG, "Running u8 tests...");
+    //esp_nn_add_elementwise_u8_test();
+    //esp_nn_depthwise_conv_u8_test();
+    //esp_nn_conv_u8_test();
+    //esp_nn_avg_pool_u8_test();
+    //esp_nn_max_pool_u8_test();
+    //esp_nn_fully_connected_u8_test();
+    //ESP_LOGI(TAG, "u8 tests done!\n");
+}
--- a/code/components/esp-nn/test_app/sdkconfig.defaults
+++ b/code/components/esp-nn/test_app/sdkconfig.defaults
@@ -0,0 +1,5 @@
+
+#
+# esp-nn
+#
+CONFIG_NN_ESP32=y
--- a/code/components/esp-nn/tests/CMakeLists.txt
+++ b/code/components/esp-nn/tests/CMakeLists.txt
@@ -0,0 +1,15 @@
+
+set(COMPONENT_ADD_INCLUDEDIRS ./include/)
+set(COMPONENT_SRCS "src/basic_math_test.c"
+                   "src/convolution_test.c"
+                   "src/fully_connected_test.c"
+                   "src/pooling_test.c"
+                   "src/relu_test.c"
+                   "src/softmax_test.c")
+
+set(COMPONENT_REQUIRES )
+set(COMPONENT_PRIV_REQUIRES esp-nn)
+
+register_component()
+
+target_compile_options(${COMPONENT_LIB} PRIVATE -Wno-unused-function)
--- a/code/components/esp-nn/tests/README.md
+++ b/code/components/esp-nn/tests/README.md
@@ -0,0 +1,4 @@
+# Tests for esp_nn library
+
+- Include these in your test framework and run the framework.
+- For IDF test please refer `test_app`
--- a/code/components/esp-nn/tests/component.mk
+++ b/code/components/esp-nn/tests/component.mk
@@ -0,0 +1,5 @@
+#FIXME
+
+COMPONENT_ADD_INCLUDEDIRS := include/
+
+COMPONENT_SRCDIRS :=  src/
--- a/code/components/esp-nn/tests/include/test_functions.h
+++ b/code/components/esp-nn/tests/include/test_functions.h
@@ -0,0 +1,48 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+
+/* int8_t ops tests */
+void esp_nn_add_elementwise_s8_test();
+void esp_nn_mul_elementwise_s8_test();
+
+void esp_nn_depthwise_conv_s8_test();
+void esp_nn_conv_s8_test();
+
+void esp_nn_avg_pool_s8_test();
+void esp_nn_max_pool_s8_test();
+
+void esp_nn_fully_connected_s8_test();
+
+void esp_nn_relu6_s8_test();
+
+void esp_nn_softmax_s8_test();
+
+/* uint8_t ops tests */
+void esp_nn_add_elementwise_u8_test();
+
+void esp_nn_depthwise_conv_u8_test();
+void esp_nn_conv_u8_test();
+
+void esp_nn_avg_pool_u8_test();
+void esp_nn_max_pool_u8_test();
+
+void esp_nn_fully_connected_u8_test();
+
+/* instructions test functions */
+void compare_instructions_test();
+void arith_instructions_test();
+void min_max_instructions_test();
+void bitwise_instructions_test();
+void load_store_instructions_test();
--- a/code/components/esp-nn/tests/include/test_utils.h
+++ b/code/components/esp-nn/tests/include/test_utils.h
@@ -0,0 +1,87 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+#include <stdbool.h>
+#include <common_functions.h>
+#include <stdio.h>
+
+/* mult value range */
+#define MULT_MAX    INT32_MAX
+#define MULT_MIN    0
+
+/* shift value range */
+#define SHIFT_MIN   -31
+#define SHIFT_MAX   30
+
+/**
+ * @brief callback function to run before C function
+ */
+void profile_c_start();
+
+/**
+ * @brief callback function to run after C function
+ */
+void profile_c_end();
+
+/**
+ * @brief callback function to run before optimized function
+ */
+void profile_opt_start();
+
+/**
+ * @brief callback function to run after optimized function
+ */
+void profile_opt_end();
+
+#define ANSI_COLOR_RED     "\x1b[31m"
+#define ANSI_COLOR_GREEN   "\x1b[32m"
+#define ANSI_COLOR_YELLOW  "\x1b[33m"
+#define ANSI_COLOR_BLUE    "\x1b[34m"
+#define ANSI_COLOR_MAGENTA "\x1b[35m"
+#define ANSI_COLOR_CYAN    "\x1b[36m"
+#define ANSI_COLOR_RESET   "\x1b[0m"
+
+#define CHECK_EQUAL(ARRAY1, ARRAY2, size) ({    \
+    bool res = true;                            \
+    for (int _i = 0; _i < size; _i++) {         \
+        if (ARRAY1[_i] != ARRAY2[_i]) {         \
+            res = false;                        \
+            break;                              \
+        }                                       \
+    }                                           \
+    res;                                        \
+})
+
+#define PRINT_ARRAY_INT(ARRAY, width, height) ({        \
+    int *_array = (int *) ARRAY;                        \
+    for (int _j = 0; _j < height; _j++) {               \
+        for (int _i = 0; _i < width; _i++) {            \
+            printf("%d\t", _array[width * _j + _i]);    \
+        }                                               \
+        printf("\n");                                   \
+    }                                                   \
+    printf("\n");                                       \
+})
+
+#define PRINT_ARRAY_HEX(ARRAY, width, height) ({        \
+    uint8_t *_array = (uint8_t *) ARRAY;                \
+    for (int _j = 0; _j < height; _j++) {               \
+        for (int _i = 0; _i < width; _i++) {            \
+            printf("%02x\t", _array[width * _j + _i]);  \
+        }                                               \
+        printf("\n");                                   \
+    }                                                   \
+    printf("\n");                                       \
+})
--- a/code/components/esp-nn/tests/src/basic_math_test.c
+++ b/code/components/esp-nn/tests/src/basic_math_test.c
@@ -0,0 +1,343 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <malloc.h>
+
+#include <common_functions.h>
+#include <esp_nn.h>
+#include "test_utils.h"
+
+#if CONFIG_IDF_CMAKE
+#define IDF_HEAP_CAPS 1
+
+#if IDF_HEAP_CAPS
+#include "esp_heap_caps.h"
+#endif
+#endif
+
+void esp_nn_add_elementwise_s8_test()
+{
+    /* prepare data */
+    const int size = 1600 + 8 + 7; /* odd len to test leftover */
+    int8_t *input1;
+    int8_t *input2;
+    int8_t *out_data_c;
+    int8_t *out_data_opt;
+    int8_t *input1_orig = NULL;
+    int8_t *input2_orig = NULL;
+    int8_t *out_c_orig = NULL;
+    int8_t *out_opt_orig = NULL;
+    int32_t input1_offset = 34;
+    int32_t input2_offset = 35;
+    int32_t output_offset = 36;
+    int32_t input1_shift = -8; // right_shift amt always <= 0
+    int32_t input2_shift = -8; // right_shift amt always <= 0
+    int32_t output_shift = -9; // right_shift amt always <= 0
+    int32_t left_shift = 15; // always +ve
+    int32_t input1_mult = INT32_MAX;
+    int32_t input2_mult = INT32_MAX;
+    int32_t output_mult = INT32_MAX;
+    int32_t activation_min = -128;
+    int32_t activation_max = 127;
+
+    for (int itr = 0; itr < 10; itr++) {
+        switch (itr) {
+        case 0: // all zeros
+            input1_offset = 0;
+            input2_offset = 0;
+            output_offset = 0;
+            input1_mult = 0;
+            input2_mult = 0;
+            output_mult = 0;
+            input1_shift = 0;
+            input2_shift = 0;
+            output_shift = 0;
+            left_shift = 0;
+        break;
+        case 1: // hit min
+            input1_offset = -127;
+            input2_offset = -127;
+            output_offset = -128;
+            input1_mult = MULT_MIN;
+            input2_mult = MULT_MIN;
+            output_mult = MULT_MIN;
+            input1_shift = 0;
+            input2_shift = 0;
+            output_shift = 0;
+            left_shift = 0;
+        break;
+        case 2: // hit max
+            input1_offset = 128;
+            input2_offset = 128;
+            output_offset = -127;
+            input1_mult = MULT_MAX;
+            input2_mult = MULT_MAX;
+            output_mult = MULT_MAX;
+            input1_shift = SHIFT_MIN;
+            input2_shift = SHIFT_MIN;
+            output_shift = SHIFT_MIN;
+            left_shift = 30 - 8; // since input is 8 bits
+        break;
+        case 3: // hit extreme max
+            input1_offset = 128;
+            input2_offset = 128;
+            output_offset = -127;
+            input1_mult = MULT_MAX;
+            input2_mult = MULT_MAX;
+            output_mult = MULT_MAX;
+            input1_shift = 0;
+            input2_shift = 0;
+            output_shift = 0;
+            left_shift = 30 - 8; // -8 since input is 8 bit
+        break;
+        default:  // practical random input
+            input1_offset = rand() % 256 - 127; // range [-127, 128]
+            input2_offset = rand() % 256 - 127; // range [-127, 128]
+            output_offset = rand() % 256 - 128; // range [-128, 127]
+            input1_mult = MULT_MAX / 2 + rand() % INT16_MAX;
+            input2_mult = MULT_MAX / 2 + rand() % INT16_MAX;
+            output_mult = MULT_MAX / 2 + rand() % INT16_MAX;
+            input1_shift = -8 + rand() % 4;
+            input2_shift = -8 + rand() % 4;
+            output_shift = -8 + rand() % 4;
+            left_shift = rand() % 15;
+        }
+#if IDF_HEAP_CAPS
+        input1_orig = (int8_t *) heap_caps_malloc(size + 32, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+        input2_orig = (int8_t *) heap_caps_malloc(size + 32, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+        out_c_orig = (int8_t *) heap_caps_malloc(size + 32, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+        out_opt_orig = (int8_t *) heap_caps_malloc(size + 32, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+
+        input1 = 16 + input1_orig - ((uint32_t) input1_orig & 0xf);
+        input2 = 16 + input2_orig - ((uint32_t) input2_orig & 0xf);
+        out_data_c = 16 + out_c_orig - ((uint32_t) out_c_orig & 0xf);
+        out_data_opt = 16 + out_opt_orig - ((uint32_t) out_opt_orig & 0xf);
+#else
+        input1 = memalign(16, size);
+        input2 = memalign(16, size);
+        out_data_c = memalign(16, size);
+        out_data_opt = memalign(16, size);
+
+        input1_orig = input1;
+        input2_orig = input2;
+        out_c_orig = out_data_c;
+        out_opt_orig = out_data_opt;
+#endif
+
+        for (int i = 0; i < size; ++i) {
+            input1[i] = rand() % 256 - 128;
+            input2[i] = rand() % 256 - 128;
+        }
+
+        if (itr == 0) {
+            /* enable profiler */
+            profile_c_start();
+        }
+        /* C function */
+        esp_nn_add_elementwise_s8_ansi(input1, input2, input1_offset, input2_offset,
+                                       input1_mult, input2_mult, input1_shift, input2_shift,
+                                       left_shift, out_data_c, output_offset, output_mult,
+                                       output_shift, activation_min, activation_max, size);
+
+        if (itr == 0) {
+            profile_c_end();
+            profile_opt_start();
+        }
+
+        /* Optimized function */
+        esp_nn_add_elementwise_s8(input1, input2, input1_offset, input2_offset,
+                                  input1_mult, input2_mult, input1_shift, input2_shift,
+                                  left_shift, out_data_opt, output_offset, output_mult,
+                                  output_shift, activation_min, activation_max, size);
+        if (itr == 0) {
+            /* disable profiler */
+            profile_opt_end();
+        }
+
+        bool ret = CHECK_EQUAL(out_data_c, out_data_opt, size);
+        if (ret == false) {
+            printf(ANSI_COLOR_RED"%s[%d] failed\n"ANSI_COLOR_RESET, __FUNCTION__, itr);
+            printf("Output: \n");
+            PRINT_ARRAY_HEX(out_data_opt, size, 1);
+            printf("Expected: \n");
+            PRINT_ARRAY_HEX(out_data_c, size, 1);
+            printf("Input1:\n");
+            PRINT_ARRAY_HEX(input1, size, 1);
+            printf("Input2:\n");
+            PRINT_ARRAY_HEX(input2, size, 1);
+            printf("in1_shift %d, in2_shift %d, left_shift %d, out_shift %d\n",
+                   input1_shift, input2_shift, left_shift, output_shift);
+            printf("in1_mult %d, in2_mult %d, out_mult %d\n", input1_mult, input2_mult, output_mult);
+            goto elementwise_add_test_cleanup;
+        }
+        printf(ANSI_COLOR_GREEN"%s[%d] passed\n"ANSI_COLOR_RESET, __FUNCTION__, itr);
+
+elementwise_add_test_cleanup:
+        if (input1_orig) {
+            free(input1_orig);
+        }
+        if (input2_orig) {
+            free(input2_orig);
+        }
+        if (out_data_c) {
+            free(out_c_orig);
+        }
+        if (out_data_opt) {
+            free(out_opt_orig);
+        }
+    }
+}
+
+void esp_nn_mul_elementwise_s8_test()
+{
+    /* prepare data */
+    const int size = 1600 + 8 + 7; /* odd len to test leftover */
+    int8_t *input1;
+    int8_t *input2;
+    int8_t *out_data_c;
+    int8_t *out_data_opt;
+    int32_t input1_offset = 34;
+    int32_t input2_offset = 35;
+    int32_t output_offset = 36;
+    int32_t output_shift = -7;
+    int32_t output_mult = MULT_MAX; // max out_mult
+    int32_t activation_min = -128;
+    int32_t activation_max = 127;
+    int8_t *input1_orig = NULL;
+    int8_t *input2_orig = NULL;
+    int8_t *out_c_orig = NULL;
+    int8_t *out_opt_orig = NULL;
+
+    for (int itr = 0; itr < 10; itr++) {
+        switch (itr) {
+        case 0: // all zeros
+            input1_offset = 0;
+            input2_offset = 0;
+            output_offset = 0;
+            output_mult = 0;
+            output_shift = 0;
+        break;
+        case 1: // hit min
+            input1_offset = -127;
+            input2_offset = -127;
+            output_offset = -128;
+            output_mult = MULT_MIN;
+            output_shift = 0;
+        break;
+        case 2: // hit max
+            input1_offset = 128;
+            input2_offset = 128;
+            output_offset = -127;
+            output_mult = MULT_MAX;
+            output_shift = SHIFT_MIN;
+        break;
+        case 3: // hit extreme max
+            input1_offset = 128;
+            input2_offset = 128;
+            output_offset = -127;
+            output_mult = MULT_MAX;
+            output_shift = 0;
+        break;
+        default:  // practical random input
+            input1_offset = rand() % 256 - 127; // range [-127, 128]
+            input2_offset = rand() % 256 - 127; // range [-127, 128]
+            output_offset = rand() % 256 - 128; // range [-128, 127]
+            output_mult = MULT_MAX / 2 + rand() % INT16_MAX;
+            output_shift = -8 + rand() % 4;
+        }
+
+#if IDF_HEAP_CAPS
+        input1_orig = (int8_t *) heap_caps_malloc(size + 32, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+        input2_orig = (int8_t *) heap_caps_malloc(size + 32, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+        out_c_orig = (int8_t *) heap_caps_malloc(size + 32, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+        out_opt_orig = (int8_t *) heap_caps_malloc(size + 32, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+
+        input1 = 16 + input1_orig - ((uint32_t) input1_orig & 0xf);
+        input2 = 16 + input2_orig - ((uint32_t) input2_orig & 0xf);
+        out_data_c = 16 + out_c_orig - ((uint32_t) out_c_orig & 0xf);
+        out_data_opt = 16 + out_opt_orig - ((uint32_t) out_opt_orig & 0xf);
+#else
+        input1 = memalign(16, size);
+        input2 = memalign(16, size);
+        out_data_c = memalign(16, size);
+        out_data_opt = memalign(16, size);
+
+        input1_orig = input1;
+        input2_orig = input2;
+        out_c_orig = out_data_c;
+        out_opt_orig = out_data_opt;
+#endif
+
+        for (int i = 0; i < size; ++i) {
+            input1[i] = rand() % 256 - 128;
+            input2[i] = rand() % 256 - 128;
+        }
+
+        if (itr == 0) {
+            /* enable profiler */
+            profile_c_start();
+        }
+        /* C function */
+        esp_nn_mul_elementwise_s8_ansi(input1, input2, input1_offset, input2_offset,
+                                       out_data_c, output_offset, output_mult, output_shift,
+                                       activation_min, activation_max, size);
+
+        if (itr == 0) {
+            profile_c_end();
+            profile_opt_start();
+        }
+        /* Optimized function */
+        esp_nn_mul_elementwise_s8(input1, input2, input1_offset, input2_offset,
+                                  out_data_opt, output_offset, output_mult, output_shift,
+                                  activation_min, activation_max, size);
+
+        if (itr == 0) {
+            /* disable profiler */
+            profile_opt_end();
+        }
+
+        bool ret = CHECK_EQUAL(out_data_c, out_data_opt, size);
+        if (ret == false) {
+            printf(ANSI_COLOR_RED"%s[%d] failed\n"ANSI_COLOR_RESET, __FUNCTION__, itr);
+            printf("Output: \n");
+            PRINT_ARRAY_HEX(out_data_opt, size, 1);
+            printf("Expected: \n");
+            PRINT_ARRAY_HEX(out_data_c, size, 1);
+            printf("Input1:\n");
+            PRINT_ARRAY_HEX(input1, size, 1);
+            printf("Input2:\n");
+            PRINT_ARRAY_HEX(input2, size, 1);
+            goto elementwise_mult_test_cleanup;
+        }
+        printf(ANSI_COLOR_GREEN"%s[%d] passed\n"ANSI_COLOR_RESET, __FUNCTION__, itr);
+
+elementwise_mult_test_cleanup:
+        if (input1_orig) {
+            free(input1_orig);
+        }
+        if (input2_orig) {
+            free(input2_orig);
+        }
+        if (out_data_c) {
+            free(out_c_orig);
+        }
+        if (out_data_opt) {
+            free(out_opt_orig);
+        }
+    }
+}
--- a/code/components/esp-nn/tests/src/convolution_test.c
+++ b/code/components/esp-nn/tests/src/convolution_test.c
@@ -0,0 +1,571 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <malloc.h>
+
+#include <esp_nn.h>
+#include "test_utils.h"
+
+#if CONFIG_IDF_CMAKE
+#define IDF_HEAP_CAPS 1
+
+#if IDF_HEAP_CAPS
+#include "esp_heap_caps.h"
+#endif
+#endif
+
+void esp_nn_depthwise_conv_s8_test()
+{
+    int8_t *input = NULL, *filter_data = NULL, *out_data_c = NULL, *out_data_opt = NULL;
+    int32_t *bias = NULL;
+    int32_t input_offset = 5; /* some number in [-128, 127] */
+    int32_t out_offset = 7;
+    int32_t activation_min = -125;
+    int32_t activation_max = 120;
+    void *scratch_buf = NULL;
+
+    /* independent variables */
+    int input_wd, input_ht, channels;
+    uint16_t filter_ht, filter_wd, ch_mult;
+    uint16_t pad_wd, pad_ht, stride_wd, stride_ht;
+
+    // run for 10 iterations
+    for (int itr = 0; itr < 10; itr++) {
+        /* prepare data */
+        switch (itr) {
+        case 0: // (ch_mult 1, (channels % 16) = 0), filter (3,3), pad (0,0)
+            input_wd = 18;
+            input_ht = 18;
+            filter_ht = 3;
+            filter_wd = 3;
+            ch_mult = 1;
+            channels = 16;
+            pad_wd = 0;
+            pad_ht = 0;
+            stride_wd = 1;
+            stride_ht = 1;
+            break;
+        case 1: // (ch_mult 1, (channels % 16) = 0), filter (3,3), pad (1,1)
+            input_wd = 10;
+            input_ht = 10;
+            filter_ht = 3;
+            filter_wd = 3;
+            ch_mult = 1;
+            channels = 16;
+            pad_wd = 1;
+            pad_ht = 1;
+            stride_wd = 1;
+            stride_ht = 1;
+            break;
+        case 2: // (ch_mult 1, (channels % 8) = 0), filter (3,3), pad (1,1)
+            input_wd = 10;
+            input_ht = 10;
+            filter_ht = 3;
+            filter_wd = 3;
+            ch_mult = 1;
+            channels = 24;
+            pad_wd = 1;
+            pad_ht = 1;
+            stride_wd = 1;
+            stride_ht = 1;
+            break;
+        case 3: // other filter sizes (ch_mult 1, (channels % 8) = 0)
+            input_wd = 10;
+            input_ht = 10;
+            filter_ht = 3;
+            filter_wd = 3;
+            ch_mult = 1;
+            channels = 24;
+            pad_wd = 1;
+            pad_ht = 1;
+            stride_wd = 1;
+            stride_ht = 1;
+            break;
+        case 4: // other filter sizes (ch_mult 8 = 0)
+            input_wd = 6;
+            input_ht = 6;
+            filter_ht = 3;
+            filter_wd = 3;
+            ch_mult = 8;
+            channels = 4;
+            pad_wd = 1;
+            pad_ht = 1;
+            stride_wd = 1;
+            stride_ht = 1;
+            break;
+        case 5: // other filter sizes (ch_mult 8 = 0)
+            input_wd = 12;
+            input_ht = 12;
+            filter_ht = 5;
+            filter_wd = 5;
+            ch_mult = 8;
+            channels = 4;
+            pad_wd = 1;
+            pad_ht = 1;
+            stride_wd = 1;
+            stride_ht = 1;
+            break;
+        case 6: // other filter sizes (ch_mult 4 = 0)
+            input_wd = 6;
+            input_ht = 6;
+            filter_ht = 3;
+            filter_wd = 3;
+            ch_mult = 4;
+            channels = 4;
+            pad_wd = 1;
+            pad_ht = 1;
+            stride_wd = 1;
+            stride_ht = 1;
+            break;
+        case 7: // (ch_mult 1, (channels % 16) = 0), filter (3,3), pad (0,0)  stride (2,2)
+            input_wd = 6;
+            input_ht = 6;
+            filter_ht = 3;
+            filter_wd = 3;
+            ch_mult = 1;
+            channels = 16;
+            pad_wd = 0;
+            pad_ht = 0;
+            stride_wd = 2;
+            stride_ht = 2;
+            break;
+        default:
+            input_wd = 4;
+            input_ht = 4;
+            filter_ht = 3;
+            filter_wd = 3;
+            ch_mult = 4;
+            channels = 4;
+            pad_wd = 1;
+            pad_ht = 1;
+            stride_wd = 1;
+            stride_ht = 1;
+            break;
+        }
+
+        uint16_t out_wd = (input_wd - filter_wd + 1) / stride_wd;
+        uint16_t out_ht = (input_ht - filter_ht + 1) / stride_ht;
+        int in_size = input_wd * input_ht * channels;
+        int out_size = out_wd * out_ht * channels * ch_mult;
+        int filter_size = filter_wd * filter_ht * channels * ch_mult + 4;
+        int bias_size = channels * ch_mult + 1;
+        int32_t out_shift[channels * ch_mult];
+        int32_t out_mult[channels * ch_mult];
+
+#if IDF_HEAP_CAPS
+        int8_t *input_orig = (int8_t *) heap_caps_malloc(in_size + 32, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+        int8_t *out_c_orig = (int8_t *) heap_caps_malloc(out_size + 32, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+        int8_t *out_opt_orig = (int8_t *) heap_caps_malloc(out_size + 32, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+        filter_data = (int8_t *) heap_caps_malloc(filter_size, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+        bias = (int32_t *) heap_caps_malloc(bias_size * 4, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+
+        input = 16 + input_orig - ((uint32_t) input_orig & 0xf);
+        out_data_c = 16 + out_c_orig - ((uint32_t) out_c_orig & 0xf);
+        out_data_opt = 16 + out_opt_orig - ((uint32_t) out_opt_orig & 0xf);
+#else
+        input = memalign(16, in_size + 16);
+        filter_data = memalign(16, filter_size);
+        out_data_c = memalign(16, out_size + 16);
+        out_data_opt = memalign(16, out_size + 16);
+        bias = memalign(16, bias_size * 4);
+        int8_t *input_orig = input;
+        int8_t *out_c_orig = out_data_c;
+        int8_t *out_opt_orig = out_data_opt;
+#endif
+        if (bias == NULL || input == NULL || filter_data == NULL ||
+                out_data_c == NULL || out_data_opt == NULL || bias == NULL) {
+            printf(ANSI_COLOR_RED"%s[%d] allocations failed\n"ANSI_COLOR_RESET, __FUNCTION__, itr);
+            goto dc_s8_cleanup;
+        }
+
+        /* Generate input data */
+        for (int i = 0; i < in_size; ++i) {
+            input[i] = rand() % 128;
+        }
+
+        /* Generate filter data */
+        for (int i = 0; i < filter_size; ++i) {
+            filter_data[i] = rand() % 256 - 128;
+        }
+
+        /* Generate bias data */
+        for (int i = 0; i < channels * ch_mult; ++i) {
+            bias[i + 1] = rand() % INT16_MAX; //0th index left for unalignment
+            out_shift[i] = -8 + rand() % 3;
+            out_mult[i] = 0x7eb0e200 + rand() % 50;
+        }
+
+        int scratch_buf_size = esp_nn_get_depthwise_conv_scratch_size(input_wd, input_ht,
+                                                                    channels, ch_mult,
+                                                                    filter_wd, filter_ht);
+        if (scratch_buf_size > 0) {
+#if IDF_HEAP_CAPS
+            scratch_buf = heap_caps_malloc(scratch_buf_size + 32, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+            int align_sz = 16 - (((int32_t) scratch_buf) & 0xf);
+#else
+            scratch_buf = memalign(16, scratch_buf_size);
+            int align_sz = 0;
+#endif
+            if (scratch_buf == NULL) {
+                printf(ANSI_COLOR_RED"%s[%d] scratch_buf alloc failed size %d\n"ANSI_COLOR_RESET,
+                       __FUNCTION__, itr, scratch_buf_size);
+                goto dc_s8_cleanup;
+            }
+            esp_nn_set_depthwise_conv_scratch_buf(scratch_buf + align_sz);
+        }
+        if (itr == 0) {
+            /* enable profiler */
+            profile_c_start();
+        }
+
+        /* C function */
+        esp_nn_depthwise_conv_s8_ansi(input, input_wd, input_ht, channels, input_offset,
+                                    pad_wd, pad_ht, stride_wd, stride_ht, ch_mult,
+                                    filter_data + 4, filter_wd, filter_ht,
+                                    bias + 1, out_data_c, out_wd, out_ht, out_offset, out_shift,
+                                    out_mult, activation_min, activation_max);
+
+        if (itr == 0) {
+            profile_c_end();
+            profile_opt_start();
+        }
+
+        /* Optimized function */
+        esp_nn_depthwise_conv_s8(input, input_wd, input_ht, channels, input_offset,
+                                pad_wd, pad_ht, stride_wd, stride_ht, ch_mult,
+                                filter_data + 4, filter_wd, filter_ht,
+                                bias + 1, out_data_opt, out_wd, out_ht, out_offset, out_shift,
+                                out_mult, activation_min, activation_max);
+
+        if (itr == 0) {
+            /* disable profiler */
+            profile_opt_end();
+        }
+
+        bool ret = CHECK_EQUAL(out_data_c, out_data_opt, out_size);
+        if (ret == false) {
+            printf(ANSI_COLOR_RED"%s[%d] failed\n"ANSI_COLOR_RESET, __FUNCTION__, itr);
+            printf("Output: \n");
+            PRINT_ARRAY_HEX(out_data_opt, out_size / out_ht, out_ht);
+            printf("Expected: \n");
+            PRINT_ARRAY_HEX(out_data_c, out_size / out_ht, out_ht);
+            printf("Input:\n");
+            PRINT_ARRAY_HEX(input, in_size / input_ht, input_ht);
+            printf("Filter data:\n");
+            PRINT_ARRAY_HEX(filter_data + 4, (filter_size - 4) / filter_ht, filter_ht);
+            printf("bias data:\n");
+            PRINT_ARRAY_INT(bias + 1, ch_mult * channels, 1);
+            goto dc_s8_cleanup;
+        }
+        printf(ANSI_COLOR_GREEN"%s[%d] passed\n"ANSI_COLOR_RESET, __FUNCTION__, itr);
+
+    dc_s8_cleanup:
+        if (input) {
+            free(input_orig);
+        }
+        if (filter_data) {
+            free(filter_data);
+        }
+        if (out_data_c) {
+            free(out_c_orig);
+        }
+        if (out_data_opt) {
+            free(out_opt_orig);
+        }
+        if (bias) {
+            free(bias);
+        }
+        if (scratch_buf) {
+            free(scratch_buf);
+        }
+    }
+}
+
+void esp_nn_conv_s8_test()
+{
+    const int32_t input_offset = 5; /* some number in [-128, 127] */
+    const int32_t activation_min = -125;
+    const int32_t activation_max = 122;
+    const int32_t out_offset = 3;
+
+    void *scratch_buf = NULL;
+    int8_t *input_orig;
+    int8_t *out_c_orig;
+    int8_t *out_opt_orig;
+    int8_t *filter_data;
+    int32_t *bias;
+
+    /* independent variable */
+    int in_wd, in_ht, in_channels, out_channels;
+    uint16_t filter_ht, filter_wd;
+    uint16_t pad_wd, pad_ht, stride_wd, stride_ht;
+
+    // run for 10 iterations
+    for (int itr = 0; itr < 10; itr++) {
+        switch (itr) {
+        case 0: // ch % 8 == 0 && filter (1,1), padding (0,0)
+            in_wd = 10;
+            in_ht = 10;
+            in_channels = 64;
+            out_channels = 64;
+            filter_ht = 1;
+            filter_wd = 1;
+            pad_wd = 0;
+            pad_ht = 0;
+            stride_wd = 1;
+            stride_ht = 1;
+            break;
+        case 1: // ch % 4 == 0 && (in_wd * in_ht) % 16 == 0
+            in_wd = 4;
+            in_ht = 4;
+            in_channels = 20;
+            out_channels = 8;
+            filter_ht = 1;
+            filter_wd = 1;
+            pad_wd = 0;
+            pad_ht = 0;
+            stride_wd = 1;
+            stride_ht = 1;
+            break;
+        case 2: // ch, filter (3x3x3)
+            in_wd = 10;
+            in_ht = 10;
+            in_channels = 3;
+            out_channels = 64;
+            filter_ht = 3;
+            filter_wd = 3;
+            pad_wd = 0;
+            pad_ht = 0;
+            stride_wd = 1;
+            stride_ht = 1;
+            break;
+        case 3: // remaining pad (0, 0)
+            in_wd = 10;
+            in_ht = 10;
+            in_channels = 3;
+            out_channels = 64;
+            filter_ht = 1;
+            filter_wd = 1;
+            pad_wd = 0;
+            pad_ht = 0;
+            stride_wd = 1;
+            stride_ht = 1;
+            break;
+        case 4: // unopt case
+            in_wd = 10;
+            in_ht = 10;
+            in_channels = 12;
+            out_channels = 64;
+            filter_ht = 3;
+            filter_wd = 3;
+            pad_wd = 1;
+            pad_ht = 1;
+            stride_wd = 1;
+            stride_ht = 1;
+            break;
+        case 5: // ch % 8 == 0 & stride (2,2)
+            in_wd = 16;
+            in_ht = 16;
+            in_channels = 16;
+            out_channels = 16;
+            filter_ht = 1;
+            filter_wd = 1;
+            pad_wd = 0;
+            pad_ht = 0;
+            stride_wd = 2;
+            stride_ht = 2;
+            break;
+        case 6: // ch % 8 == 0 && filter (1,1), padding (0,0)
+            in_wd = 2;
+            in_ht = 2;
+            in_channels = 8;
+            out_channels = 8;
+            filter_ht = 1;
+            filter_wd = 1;
+            pad_wd = 0;
+            pad_ht = 0;
+            stride_wd = 1;
+            stride_ht = 1;
+            break;
+        default: // ch % 8 == 0
+            in_wd = 8;
+            in_ht = 8;
+            in_channels = 16;
+            out_channels = 16;
+            filter_ht = 1;
+            filter_wd = 1;
+            pad_wd = 0;
+            pad_ht = 0;
+            stride_wd = 1;
+            stride_ht = 1;
+            break;
+        }
+
+        /* prepare data */
+        uint16_t out_wd = (in_wd - filter_wd + 1) / stride_wd;
+        uint16_t out_ht = (in_ht - filter_ht + 1) / stride_ht;
+
+        int in_size = in_wd * in_ht * in_channels;
+        int filter_size = filter_wd * filter_ht * in_channels * out_channels + 2;
+        int out_size = out_wd * out_ht * out_channels;
+
+#if IDF_HEAP_CAPS
+        input_orig = (int8_t *) heap_caps_malloc(in_size + 32, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+        out_c_orig = (int8_t *) heap_caps_malloc(out_size + 32, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+        out_opt_orig = (int8_t *) heap_caps_malloc(out_size + 32, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+        filter_data = (int8_t *) heap_caps_malloc(filter_size + 32, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+        bias = (int32_t *) heap_caps_malloc(128 + sizeof (int32_t) * out_channels, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+
+        int8_t *input = 16 + input_orig - ((uint32_t) input_orig & 0xf);
+        int8_t *out_data_c = 16 + out_c_orig - ((uint32_t) out_c_orig & 0xf);
+        int8_t *out_data_opt = 16 + out_opt_orig - ((uint32_t) out_opt_orig & 0xf);
+#else
+        int8_t *input = memalign(16, in_size);
+        int8_t *out_data_c = memalign(16, out_size);
+        int8_t *out_data_opt = memalign(16, out_size);
+        filter_data = memalign(16, filter_size);
+        bias = calloc(1, 128 + sizeof (int32_t) * out_channels);
+        input_orig = input;
+        out_c_orig = out_data_c;
+        out_opt_orig = out_data_opt;
+#endif
+        int32_t *out_shift = calloc(1, 128 + sizeof (int32_t) * out_channels);
+        int32_t *out_mult = calloc(1, 128 + sizeof (int32_t) * out_channels);
+
+        if (input == NULL || filter_data == NULL ||
+                out_data_c == NULL || out_data_opt == NULL) {
+            printf(ANSI_COLOR_RED"%s allocations failed\n"ANSI_COLOR_RESET, __FUNCTION__);
+            goto conv_s8_cleanup;
+        }
+
+        if (bias == NULL || out_shift == NULL || out_mult == NULL) {
+            printf(ANSI_COLOR_RED"%s allocations failed\n"ANSI_COLOR_RESET, __FUNCTION__);
+            goto conv_s8_cleanup;
+        }
+
+        /* Generate input data between -128 -> +127 */
+        for (int i = 0; i < in_size; ++i) {
+            input[i] = rand() % 255 - 128;
+        }
+
+        /* Generate filter data between -128 -> +127 */
+        for (int i = 0; i < filter_size; ++i) {
+            filter_data[i] = rand() % 256 - 128;
+        }
+
+        /* Generate bias data */
+        for (int i = 0; i < out_channels; ++i) {
+            bias[i] = (int32_t)rand() % UINT16_MAX + UINT8_MAX;
+        }
+
+        /* Shift and multiplier */
+        for (int i = 0; i < out_channels; ++i) {
+            out_shift[i] = -10 + rand() % 2;
+            out_mult[i] = 0x7f67f4f8 + rand() % 50;
+        }
+
+        int scratch_buf_size = esp_nn_get_conv_scratch_size(in_wd, in_ht, in_channels,
+                                                            out_channels, filter_wd, filter_ht);
+        if (scratch_buf_size > 0) {
+#if IDF_HEAP_CAPS
+            void *scratch_buf = heap_caps_malloc(scratch_buf_size + 32, MALLOC_CAP_SPIRAM | MALLOC_CAP_8BIT);
+            int align_sz = 16 - (((int32_t) scratch_buf) & 0xf);
+#else
+            void *scratch_buf = memalign(16, scratch_buf_size);
+            int align_sz = 0;
+#endif
+            if (scratch_buf == NULL) {
+                printf(ANSI_COLOR_RED"%s scratch_buf alloc failed size %d\n"ANSI_COLOR_RESET, __FUNCTION__, scratch_buf_size);
+                goto conv_s8_cleanup;
+            }
+            esp_nn_set_conv_scratch_buf(scratch_buf + align_sz);
+        }
+
+        if (itr == 0) {
+            /* enable profiler */
+            profile_c_start();
+        }
+
+        /* C function */
+        esp_nn_conv_s8_ansi(input, in_wd, in_ht, in_channels, input_offset,
+                            pad_wd, pad_ht, stride_wd, stride_ht,
+                            filter_data + 2, filter_wd, filter_ht, bias,
+                            out_data_c, out_wd, out_ht, out_channels, out_offset, out_shift,
+                            out_mult, activation_min, activation_max);
+
+        if (itr == 0) {
+            profile_c_end();
+            profile_opt_start();
+        }
+
+        /* Optimized function */
+        esp_nn_conv_s8(input, in_wd, in_ht, in_channels, input_offset,
+                    pad_wd, pad_ht, stride_wd, stride_ht,
+                    filter_data + 2, filter_wd, filter_ht, bias,
+                    out_data_opt, out_wd, out_ht, out_channels, out_offset, out_shift,
+                    out_mult, activation_min, activation_max);
+
+        if (itr == 0) {
+            /* disable profiler */
+            profile_opt_end();
+        }
+
+        bool ret = CHECK_EQUAL(out_data_c, out_data_opt, out_size);
+        if (ret == false) {
+            printf(ANSI_COLOR_RED"%s[%d] failed\n"ANSI_COLOR_RESET, __FUNCTION__, itr);
+            printf("Output: \n");
+            PRINT_ARRAY_HEX(out_data_opt, out_size / out_ht, out_ht);
+            printf("Expected: \n");
+            PRINT_ARRAY_HEX(out_data_c, out_size / out_ht, out_ht);
+            printf("Input:\n");
+            PRINT_ARRAY_HEX(input, in_size / in_ht, in_ht);
+            printf("Filter data:\n");
+            PRINT_ARRAY_HEX(filter_data + 2, (filter_size - 2) / filter_ht, filter_ht);
+            printf("bias data:\n");
+            PRINT_ARRAY_INT(bias, out_channels, 1);
+            goto conv_s8_cleanup;
+        }
+        printf(ANSI_COLOR_GREEN"%s[%d] passed\n"ANSI_COLOR_RESET, __FUNCTION__, itr);
+
+    conv_s8_cleanup:
+        if (input) {
+            free(input_orig);
+        }
+        if (filter_data) {
+            free(filter_data);
+        }
+        if (out_data_c) {
+            free(out_c_orig);
+        }
+        if (out_data_opt) {
+            free(out_opt_orig);
+        }
+        if (bias) {
+            free(bias);
+        }
+        if (out_shift) {
+            free(out_shift);
+        }
+        if (out_mult) {
+            free(out_mult);
+        }
+        if (scratch_buf) {
+            free(scratch_buf);
+        }
+    }
+}
--- a/code/components/esp-nn/tests/src/fully_connected_test.c
+++ b/code/components/esp-nn/tests/src/fully_connected_test.c
@@ -0,0 +1,111 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+
+#include <esp_nn.h>
+#include "test_utils.h"
+
+
+void esp_nn_fully_connected_s8_test()
+{
+    /* prepare data */
+    static uint16_t row_len = 256 + 8 + 7; /* odd len to test unaligned+left-over */
+    static uint16_t out_channels = 3;
+    int8_t input[row_len];
+    int8_t filter_data[row_len * out_channels];
+    int8_t output_c[out_channels], output_opt[out_channels];
+    static int32_t activation_min = -128;
+    static int32_t activation_max = 127;
+    static int32_t input_offset = 0;
+    static int32_t filter_offset = 0;
+    int32_t out_shift = -10;
+    static int32_t out_offset = 127;
+    int32_t out_mult = 0x59e492c4;
+    for (int itr = 0; itr < 5; itr++) {
+        out_mult = INT32_MAX / row_len + rand() % INT16_MAX;
+        switch (itr) {
+        case 0:
+            out_shift = -10;
+            break;
+        case 1:
+            out_shift = SHIFT_MIN;
+            break;
+        case 2:
+            out_shift = SHIFT_MAX;
+            break;
+        case 3:
+            out_shift = 0;
+            break;
+        default:
+            out_shift = -10 + rand() % 5;
+            break;
+        }
+        if (itr == 0) {
+            out_shift = SHIFT_MAX;
+        }
+        /* Generate input and filter data */
+        for (int i = 0; i < row_len; ++i) {
+            input[i] = rand() % 256 - 128;
+        }
+        for (int i = 0; i < row_len * out_channels; ++i) {
+            filter_data[i] = rand() % 256 - 128;
+        }
+
+        if (itr == 0) {
+            /* enable profiler */
+            profile_c_start();
+        }
+
+        /* C function */
+        esp_nn_fully_connected_s8_ansi(input, input_offset, row_len, filter_data, filter_offset,
+                                    NULL, output_c, out_channels, out_offset, out_shift, out_mult,
+                                    activation_min, activation_max);
+
+        if (itr == 0) {
+            profile_c_end();
+            profile_opt_start();
+        }
+
+        /* Optimized function */
+        esp_nn_fully_connected_s8(input, input_offset, row_len, filter_data, filter_offset,
+                                NULL, output_opt, out_channels, out_offset, out_shift, out_mult,
+                                activation_min, activation_max);
+
+        if (itr == 0) {
+            /* disable profiler */
+            profile_opt_end();
+        }
+
+        bool ret = CHECK_EQUAL(output_c, output_opt, out_channels);
+        if (ret == false) {
+            printf(ANSI_COLOR_RED"%s[%d] failed\n"ANSI_COLOR_RESET, __FUNCTION__, itr);
+            printf("Output: \n");
+            PRINT_ARRAY_HEX(output_opt, out_channels, 1);
+            printf("Expected: \n");
+            PRINT_ARRAY_HEX(output_c, out_channels, 1);
+            printf("Input:\n");
+            PRINT_ARRAY_HEX(input, row_len, 1);
+            printf("Filter data:\n");
+            PRINT_ARRAY_HEX(filter_data, row_len, out_channels);
+            printf("Out shift: %d\n", out_shift);
+            printf("Out mult: %x\n", out_mult);
+            return;
+        }
+        printf(ANSI_COLOR_GREEN"%s[%d] passed\n"ANSI_COLOR_RESET, __FUNCTION__, itr);
+    }
+}
--- a/code/components/esp-nn/tests/src/pooling_test.c
+++ b/code/components/esp-nn/tests/src/pooling_test.c
@@ -0,0 +1,184 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <malloc.h>
+
+#include <esp_nn.h>
+#include "test_utils.h"
+
+
+void esp_nn_avg_pool_s8_test()
+{
+    /* prepare data */
+    const uint16_t input_wd = 16;
+    const uint16_t input_ht = 16;
+    const uint16_t channels = 16; /* With TFLite example, I have seen it 256 */
+    const int size = input_wd * input_ht * channels;
+    int8_t *input, *output_c, *output_opt;
+    const int32_t activation_min = -128;
+    const int32_t activation_max = 127;
+    const uint16_t pad_wd = 1;
+    const uint16_t pad_ht = 1;
+    const uint16_t stride_wd = 1;
+    const uint16_t stride_ht = 1;
+    const uint16_t filter_ht = 3;
+    const uint16_t filter_wd = 3;
+    const uint16_t out_wd = input_wd / stride_wd;
+    const uint16_t out_ht = input_ht / stride_ht;
+    const int out_size = out_wd * out_ht * channels;
+
+    input = memalign(16, size);
+    output_c = memalign(16, out_size);
+    output_opt = memalign(16, out_size);
+
+    if (input == NULL || output_c == NULL || output_opt == NULL) {
+        printf(ANSI_COLOR_RED"%s allocations failed\n"ANSI_COLOR_RESET, __FUNCTION__);
+        goto avg_pool_s8_cleanup;
+    }
+    /**
+     * width/height, channels etc look suspicious but it it true.
+     * It actually depends upon where in model this is actually placed.
+     * If at the end wd/ht tends to be smaller and depth larger.
+     */
+
+    for (int i = 0; i < size; ++i) {
+        input[i] = rand() % 256 - 128;
+    }
+
+    /* enable profiler */
+    profile_c_start();
+
+    /* C function */
+    esp_nn_avg_pool_s8_ansi(input, input_wd, input_ht, output_c, out_wd, out_ht,
+                              stride_wd, stride_ht, filter_wd, filter_ht, pad_wd, pad_ht,
+                              activation_min, activation_max, channels);
+
+    profile_c_end();
+    profile_opt_start();
+
+    /* Optimized function */
+    esp_nn_avg_pool_s8(input, input_wd, input_ht, output_opt, out_wd, out_ht,
+                         stride_wd, stride_ht, filter_wd, filter_ht, pad_wd, pad_ht,
+                         activation_min, activation_max, channels);
+
+    /* disable profiler */
+    profile_opt_end();
+
+
+    bool ret = CHECK_EQUAL(output_c, output_opt, out_size);
+    if (ret == false) {
+        printf(ANSI_COLOR_RED"%s failed\n"ANSI_COLOR_RESET, __FUNCTION__);
+        printf("Output: \n");
+        PRINT_ARRAY_HEX(output_opt, out_wd * channels, out_ht);
+        printf("Expected: \n");
+        PRINT_ARRAY_HEX(output_c, out_wd * channels, out_ht);
+        printf("Input:\n");
+        PRINT_ARRAY_HEX(input, input_wd * channels, input_ht);
+        goto avg_pool_s8_cleanup;
+    }
+    printf(ANSI_COLOR_GREEN"%s passed\n"ANSI_COLOR_RESET, __FUNCTION__);
+
+avg_pool_s8_cleanup:
+    if (input) {
+        free(input);
+    }
+    if (output_c) {
+        free(output_c);
+    }
+    if (output_opt) {
+        free(output_opt);
+    }
+}
+
+void esp_nn_max_pool_s8_test()
+{
+    /* prepare data */
+    const uint16_t input_wd = 16;
+    const uint16_t input_ht = 16;
+    const uint16_t channels = 16; /* With TFLite example, I have seen it 256 */
+    int8_t *input, *output_c, *output_opt;
+    const int size = input_wd * input_ht * channels;
+    const int32_t activation_min = -128;
+    const int32_t activation_max = 127;
+    const uint16_t pad_wd = 1;
+    const uint16_t pad_ht = 1;
+    const uint16_t stride_wd = 1;
+    const uint16_t stride_ht = 1;
+    const uint16_t filter_ht = 3;
+    const uint16_t filter_wd = 3;
+    const uint16_t out_wd = input_wd / stride_wd;
+    const uint16_t out_ht = input_ht / stride_ht;
+    const int out_size = out_wd * out_ht * channels;
+
+    input = memalign(16, size);
+    output_c = memalign(16, out_size);
+    output_opt = memalign(16, out_size);
+
+    if (input == NULL || output_c == NULL || output_opt == NULL) {
+        printf(ANSI_COLOR_RED"%s allocations failed\n"ANSI_COLOR_RESET, __FUNCTION__);
+        goto max_pool_s8_cleanup;
+    }
+
+    for (int i = 0; i < size; ++i) {
+        input[i] = rand() % 256 - 128;
+    }
+
+    /* enable profiler */
+    profile_c_start();
+
+    /* C function */
+    esp_nn_max_pool_s8_ansi(input, input_wd, input_ht, output_c, out_wd, out_ht,
+                            stride_wd, stride_ht, filter_wd, filter_ht, pad_wd, pad_ht,
+                            activation_min, activation_max, channels);
+
+    profile_c_end();
+    profile_opt_start();
+
+    /* Optimized function */
+    esp_nn_max_pool_s8(input, input_wd, input_ht, output_opt, out_wd, out_ht,
+                       stride_wd, stride_ht, filter_wd, filter_ht, pad_wd, pad_ht,
+                       activation_min, activation_max, channels);
+
+    /* disable profiler */
+    profile_opt_end();
+
+
+    bool ret = CHECK_EQUAL(output_c, output_opt, out_wd * out_ht * channels);
+    if (ret == false) {
+        printf(ANSI_COLOR_RED"%s failed\n"ANSI_COLOR_RESET, __FUNCTION__);
+        printf("Output: \n");
+        PRINT_ARRAY_HEX(output_opt, out_wd * out_ht * channels, 1);
+        printf("Expected: \n");
+        PRINT_ARRAY_HEX(output_c, out_wd * out_ht * channels, 1);
+        printf("Input:\n");
+        PRINT_ARRAY_HEX(input, 8, size / 8);
+        goto max_pool_s8_cleanup;
+    }
+    printf(ANSI_COLOR_GREEN"%s passed\n"ANSI_COLOR_RESET, __FUNCTION__);
+
+max_pool_s8_cleanup:
+    if (input) {
+        free(input);
+    }
+    if (output_c) {
+        free(output_c);
+    }
+    if (output_opt) {
+        free(output_opt);
+    }
+}
--- a/code/components/esp-nn/tests/src/relu_test.c
+++ b/code/components/esp-nn/tests/src/relu_test.c
@@ -0,0 +1,83 @@
+// Copyright 2020-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <malloc.h>
+
+#include <esp_nn.h>
+#include "test_utils.h"
+
+void esp_nn_relu6_s8_test()
+{
+    const int size = 1600 + 8 + 7;
+    int8_t *input, *inout_ansi, *inout_opt;
+
+    input = memalign(16, size);
+    inout_ansi = memalign(16, size);
+    inout_opt = memalign(16, size);
+
+    if (input == NULL || inout_ansi == NULL || inout_opt == NULL) {
+        printf(ANSI_COLOR_RED"%s allocations failed\n"ANSI_COLOR_RESET, __FUNCTION__);
+        goto relu6_s8_cleanup;
+    }
+    /* Generate filter data between -128 -> +127 */
+    for (int i = 0; i < size; ++i) {
+        input[i] = rand() % 255 - 128;
+        inout_ansi[i] = input[i];
+        inout_opt[i] = input[i];
+    }
+
+    /* enable profiler */
+    profile_c_start();
+
+    /* C function */
+    esp_nn_relu6_s8_ansi(inout_ansi, size);
+
+    profile_c_end();
+    profile_opt_start();
+
+    /* Optimized function */
+    esp_nn_relu6_s8(inout_opt, size);
+
+    /* disable profiler */
+    profile_opt_end();
+
+    bool ret = CHECK_EQUAL(inout_ansi, inout_opt, size);
+    if (ret == false) {
+        printf(ANSI_COLOR_RED"%s failed\n"ANSI_COLOR_RESET, __FUNCTION__);
+        printf("Output: \n");
+        PRINT_ARRAY_HEX(inout_opt, size, 1);
+        printf("Expected: \n");
+        PRINT_ARRAY_HEX(inout_ansi, size, 1);
+        printf("Input:\n");
+        PRINT_ARRAY_HEX(input, size, 1);
+        goto relu6_s8_cleanup;
+    }
+    printf(ANSI_COLOR_GREEN"%s passed\n"ANSI_COLOR_RESET, __FUNCTION__);
+
+relu6_s8_cleanup:
+    if (input) {
+        free (input);
+    }
+    if (inout_ansi) {
+        free (inout_ansi);
+    }
+    if (inout_opt) {
+        free (inout_opt);
+    }
+
+}
--- a/code/components/esp-nn/tests/src/softmax_test.c
+++ b/code/components/esp-nn/tests/src/softmax_test.c
@@ -0,0 +1,101 @@
+// Copyright 2022 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+#include <stdbool.h>
+#include <stdio.h>
+#include <stdlib.h>
+#include <malloc.h>
+
+#include <esp_nn.h>
+#include "test_utils.h"
+
+void esp_nn_softmax_s8_test()
+{
+    const int32_t height = 8;
+    const int32_t width = 32;
+    const int32_t diff_min = -128;
+    const int32_t mult = INT32_MAX / 2;
+    const int32_t shift = 7;
+    void *scratch_buf = NULL;
+    const int size = width * height;
+    int8_t *input, *out_ansi, *out_opt;
+
+    input = memalign(16, size);
+    out_ansi = memalign(16, size);
+    out_opt = memalign(16, size);
+
+    if (input == NULL || out_ansi == NULL || out_opt == NULL) {
+        printf(ANSI_COLOR_RED"%s buffer allocations failed\n"ANSI_COLOR_RESET, __FUNCTION__);
+        goto softmax_s8_cleanup;
+    }
+
+    /* Generate input data between -128 -> +127 */
+    for (int i = 0; i < size; ++i) {
+        input[i] = rand() % 255 - 128;
+    }
+
+    /* enable profiler */
+    profile_c_start();
+
+    /* C function */
+    esp_nn_softmax_s8_ansi(input, height, width, mult, shift, diff_min, out_ansi);
+
+    profile_c_end();
+
+    int32_t scratch_buf_size = esp_nn_get_softmax_scratch_size(width, height);
+    if (scratch_buf_size) {
+        scratch_buf = memalign(4, scratch_buf_size);
+        if (scratch_buf == NULL) {
+            printf(ANSI_COLOR_RED"%s scratch_buf alloc failed size %d\n"ANSI_COLOR_RESET, __FUNCTION__, scratch_buf_size);
+            goto softmax_s8_cleanup;
+        }
+        esp_nn_set_softmax_scratch_buf(scratch_buf);
+    }
+
+    profile_opt_start();
+
+    /* Optimized function */
+    esp_nn_softmax_s8(input, height, width, mult, shift, diff_min, out_opt);
+
+    /* disable profiler */
+    profile_opt_end();
+
+    bool ret = CHECK_EQUAL(out_ansi, out_opt, size);
+    if (ret == false) {
+        printf(ANSI_COLOR_RED"%s failed\n"ANSI_COLOR_RESET, __FUNCTION__);
+        printf("Output: \n");
+        PRINT_ARRAY_HEX(out_opt, width, height);
+        printf("Expected: \n");
+        PRINT_ARRAY_HEX(out_ansi, width, height);
+        printf("Input:\n");
+        PRINT_ARRAY_HEX(input, width, height);
+        goto softmax_s8_cleanup;
+    }
+    printf(ANSI_COLOR_GREEN"%s passed\n"ANSI_COLOR_RESET, __FUNCTION__);
+
+softmax_s8_cleanup:
+    if (input) {
+        free (input);
+    }
+    if (out_ansi) {
+        free (out_ansi);
+    }
+    if (out_opt) {
+        free (out_opt);
+    }
+    if (scratch_buf) {
+        free (scratch_buf);
+    }
+}
--- a/code/components/esp-nn_20220716.zip
+++ b/code/components/esp-nn_20220716.zip
--- a/code/components/esp-nn_20220724.zip
+++ b/code/components/esp-nn_20220724.zip
--- a/code/components/esp32-camera-master.zip
+++ b/code/components/esp32-camera-master.zip
--- a/code/components/esp32-camera-master/.github/workflows/build.yml
+++ b/code/components/esp32-camera-master/.github/workflows/build.yml
@@ -0,0 +1,79 @@
+name: Build examples
+on:
+  push:
+    branches:
+    - master
+  pull_request:
+
+jobs:
+  build-master:
+    runs-on: ubuntu-latest
+    steps:
+    - name: Checkout repo
+      uses: actions/checkout@v2
+      with:
+        submodules: 'recursive'
+    - name: esp-idf build
+      uses: espressif/esp-idf-ci-action@latest
+      with:
+        path: 'examples'
+
+  build-release-v4_0:
+    runs-on: ubuntu-latest
+    steps:
+    - name: Checkout repo
+      uses: actions/checkout@v2
+      with:
+        submodules: 'recursive'
+    - name: esp-idf build
+      uses: espressif/esp-idf-ci-action@release-v4.0
+      with:
+        path: 'examples'
+
+  build-release-v4_1:
+    runs-on: ubuntu-latest
+    steps:
+    - name: Checkout repo
+      uses: actions/checkout@v2
+      with:
+        submodules: 'recursive'
+    - name: esp-idf build
+      uses: espressif/esp-idf-ci-action@release-v4.1
+      with:
+        path: 'examples'
+
+  build-release-v4_2:
+    runs-on: ubuntu-latest
+    steps:
+    - name: Checkout repo
+      uses: actions/checkout@v2
+      with:
+        submodules: 'recursive'
+    - name: esp-idf build
+      uses: espressif/esp-idf-ci-action@release-v4.2
+      with:
+        path: 'examples'
+
+  build-release-v4_3:
+    runs-on: ubuntu-latest
+    steps:
+    - name: Checkout repo
+      uses: actions/checkout@v2
+      with:
+        submodules: 'recursive'
+    - name: esp-idf build
+      uses: espressif/esp-idf-ci-action@release-v4.3
+      with:
+        path: 'examples'
+
+  build-release-v3_3:
+    runs-on: ubuntu-latest
+    steps:
+    - name: Checkout repo
+      uses: actions/checkout@v2
+      with:
+        submodules: 'recursive'
+    - name: esp-idf build
+      uses: espressif/esp-idf-ci-action@release-v3.3
+      with:
+        path: 'examples'
--- a/code/components/esp32-camera-master/.github/workflows/stale.yml
+++ b/code/components/esp32-camera-master/.github/workflows/stale.yml
@@ -0,0 +1,27 @@
+# This workflow warns and then closes issues and PRs that have had no activity for a specified amount of time.
+#
+# You can adjust the behavior by modifying this file.
+# For more information, see:
+# https://github.com/actions/stale
+name: Mark stale issues and pull requests
+
+on:
+  schedule:
+  - cron: '20 9 * * *'
+
+jobs:
+  stale:
+
+    runs-on: ubuntu-latest
+    permissions:
+      issues: write
+      pull-requests: write
+
+    steps:
+    - uses: actions/stale@v3
+      with:
+        repo-token: ${{ secrets.GITHUB_TOKEN }}
+        stale-issue-message: 'This issue appears to be stale. Please close it if its no longer valid.'
+        stale-pr-message: 'This pull request appears to be stale. Please close it if its no longer valid.'
+        stale-issue-label: 'no-issue-activity'
+        stale-pr-label: 'no-pr-activity'
--- a/code/components/esp32-camera-master/.github/workflows/upload_component.yml
+++ b/code/components/esp32-camera-master/.github/workflows/upload_component.yml
@@ -0,0 +1,21 @@
+name: Push component to https://components.espressif.com
+on:
+  push:
+    tags:
+      - v*
+jobs:
+  upload_components:
+    runs-on: ubuntu-latest
+    steps:
+      - uses: actions/checkout@master
+        with:
+          submodules: "recursive"
+
+      - name: Upload component to the component registry
+        uses: espressif/github-actions/upload_components@master
+        with:
+          name: "esp32-camera"
+          version: "git"
+          namespace: "espressif"
+          service_url: ${{ secrets.IDF_COMPONENT_API_URL }}
+          api_token: ${{ secrets.IDF_COMPONENT_API_TOKEN }}
--- a/code/components/esp32-camera-master/.gitignore
+++ b/code/components/esp32-camera-master/.gitignore
@@ -1 +1,5 @@
 *.DS_Store
+.vscode
+**/build
+**/sdkconfig
+**/sdkconfig.old
--- a/code/components/esp32-camera-master/CMakeLists.txt
+++ b/code/components/esp32-camera-master/CMakeLists.txt
@@ -1,15 +1,19 @@
-if(IDF_TARGET STREQUAL "esp32")
+if(IDF_TARGET STREQUAL "esp32" OR IDF_TARGET STREQUAL "esp32s2" OR IDF_TARGET STREQUAL "esp32s3")
  set(COMPONENT_SRCS
-    driver/camera.c
+    driver/esp_camera.c
+    driver/cam_hal.c
    driver/sccb.c
    driver/sensor.c
-    driver/xclk.c
    sensors/ov2640.c
    sensors/ov3660.c
    sensors/ov5640.c
    sensors/ov7725.c
    sensors/ov7670.c
    sensors/nt99141.c
+    sensors/gc0308.c
+    sensors/gc2145.c
+    sensors/gc032a.c
+    sensors/bf3005.c
    conversions/yuv.c
    conversions/to_jpg.cpp
    conversions/to_bmp.c
@@ -26,8 +30,34 @@ if(IDF_TARGET STREQUAL "esp32")
    driver/private_include
    sensors/private_include
    conversions/private_include
+    target/private_include
    )

+  if(IDF_TARGET STREQUAL "esp32")
+    list(APPEND COMPONENT_SRCS
+      target/xclk.c
+      target/esp32/ll_cam.c
+      )
+  endif()
+
+  if(IDF_TARGET STREQUAL "esp32s2")
+    list(APPEND COMPONENT_SRCS
+      target/xclk.c
+      target/esp32s2/ll_cam.c
+      target/esp32s2/tjpgd.c
+      )
+
+    list(APPEND COMPONENT_PRIV_INCLUDEDIRS
+      target/esp32s2/private_include
+      )
+  endif()
+
+  if(IDF_TARGET STREQUAL "esp32s3")
+    list(APPEND COMPONENT_SRCS
+      target/esp32s3/ll_cam.c
+      )
+  endif()
+
  set(COMPONENT_REQUIRES driver)
  set(COMPONENT_PRIV_REQUIRES freertos nvs_flash)

--- a/code/components/esp32-camera-master/Kconfig
+++ b/code/components/esp32-camera-master/Kconfig
@@ -5,11 +5,11 @@ menu "Camera configuration"
        default y
        help
            Enable this option if you want to use the OV7670.
-            Disable this option to safe memory.
+            Disable this option to save memory.

    config OV7725_SUPPORT
-        bool "Support OV7725 SVGA"
-        default n
+        bool "Support OV7725 VGA"
+        default y
        help
            Enable this option if you want to use the OV7725.
            Disable this option to save memory.
@@ -42,6 +42,34 @@ menu "Camera configuration"
            Enable this option if you want to use the OV5640.
            Disable this option to save memory.

+    config GC2145_SUPPORT
+        bool "Support GC2145 2MP"
+        default y
+        help
+            Enable this option if you want to use the GC2145.
+            Disable this option to save memory.
+
+    config GC032A_SUPPORT
+        bool "Support GC032A VGA"
+        default y
+        help
+            Enable this option if you want to use the GC032A.
+            Disable this option to save memory.
+
+    config GC0308_SUPPORT
+        bool "Support GC0308 VGA"
+        default y
+        help
+            Enable this option if you want to use the GC0308.
+            Disable this option to save memory.
+            
+    config BF3005_SUPPORT
+        bool "Support BF3005(BYD3005) VGA"
+        default y
+        help
+            Enable this option if you want to use the BF3005.
+            Disable this option to save memory.
+
    choice SCCB_HARDWARE_I2C_PORT
        bool "I2C peripheral to use for SCCB"
        default SCCB_HARDWARE_I2C_PORT1
@@ -53,6 +81,28 @@ menu "Camera configuration"

    endchoice

+    config SCCB_CLK_FREQ
+    int "SCCB clk frequency"
+    default 100000
+    range 100000 400000
+    help
+        Increasing this value can reduce the initialization time of the sensor.
+        Please refer to the relevant instructions of the sensor to adjust the value.
+    
+    choice GC_SENSOR_WINDOW_MODE
+        bool "GalaxyCore Sensor Window Mode"
+        depends on (GC2145_SUPPORT || GC032A_SUPPORT || GC0308_SUPPORT)
+        default GC_SENSOR_SUBSAMPLE_MODE
+        help
+            This option determines how to reduce the output size when the resolution you set is less than the maximum resolution.
+            SUBSAMPLE_MODE has a bigger perspective and WINDOWING_MODE has a higher frame rate.
+
+        config GC_SENSOR_WINDOWING_MODE
+            bool "Windowing Mode"
+        config GC_SENSOR_SUBSAMPLE_MODE
+            bool "Subsample Mode"
+    endchoice
+
    choice CAMERA_TASK_PINNED_TO_CORE
        bool "Camera task pinned to core"
        default CAMERA_CORE0
@@ -68,4 +118,12 @@ menu "Camera configuration"

    endchoice

+    config CAMERA_DMA_BUFFER_SIZE_MAX
+        int "DMA buffer size"
+        range 8192 32768
+        default 32768
+        help
+            Maximum value of DMA buffer
+            Larger values may fail to allocate due to insufficient contiguous memory blocks, and smaller value may cause DMA interrupt to be too frequent
+
 endmenu
--- a/code/components/esp32-camera-master/README.md
+++ b/code/components/esp32-camera-master/README.md
@@ -1,8 +1,30 @@
 # ESP32 Camera Driver

+[![Build examples](https://github.com/espressif/esp32-camera/actions/workflows/build.yml/badge.svg)](https://github.com/espressif/esp32-camera/actions/workflows/build.yml)
 ## General Information

-This repository hosts ESP32 compatible driver for OV2640, OV3660, OV5640, OV7670 and OV7725 image sensors. Additionally it provides a few tools, which allow converting the captured frame data to the more common BMP and JPEG formats.
+This repository hosts ESP32 series Soc compatible driver for image sensors. Additionally it provides a few tools, which allow converting the captured frame data to the more common BMP and JPEG formats.
+
+### Supported Soc
+
+- ESP32
+- ESP32-S2
+- ESP32-S3
+
+### Supported Sensor
+
+| model   | max resolution | color type | output format                                                | Len Size |
+| ------- | -------------- | ---------- | ------------------------------------------------------------ | -------- |
+| OV2640  | 1600 x 1200    | color      | YUV(422/420)/YCbCr422<br>RGB565/555<br>8-bit compressed data<br>8/10-bit Raw RGB data | 1/4"     |
+| OV3660  | 2048 x 1536    | color      | raw RGB data<br/>RGB565/555/444<br/>CCIR656<br/>YCbCr422<br/>compression | 1/5"     |
+| OV5640  | 2592 x 1944    | color      | RAW RGB<br/>RGB565/555/444<br/>CCIR656<br/>YUV422/420<br/>YCbCr422<br/>compression | 1/4"     |
+| OV7670  | 640 x 480      | color      | Raw Bayer RGB<br/>Processed Bayer RGB<br>YUV/YCbCr422<br>GRB422<br>RGB565/555 | 1/6"     |
+| OV7725  | 640 x 480      | color      | Raw RGB<br/>GRB 422<br/>RGB565/555/444<br/>YCbCr 422         | 1/4"     |
+| NT99141 | 1280 x 720     | color      | YCbCr 422<br/>RGB565/555/444<br/>Raw<br/>CCIR656<br/>JPEG compression | 1/4"     |
+| GC032A  | 640 x 480      | color      | YUV/YCbCr422<br/>RAW Bayer<br/>RGB565                        | 1/10"    |
+| GC0308  | 640 x 480      | color      | YUV/YCbCr422<br/>RAW Bayer<br/>RGB565                        | 1/6.5"   |
+| GC2145  | 1600 x 1200    | color      | YUV/YCbCr422<br/>RAW Bayer<br/>RGB565                        | 1/5"     |
+| BF3005  | 640 x 480      | color      | YUV/YCbCr422<br/>RAW Bayer<br/>RGB565                        | 1/4"     |

 ## Important to Remember

@@ -17,7 +39,7 @@ This repository hosts ESP32 compatible driver for OV2640, OV3660, OV5640, OV7670
 ### Using esp-idf

 - Clone or download and extract the repository to the components folder of your ESP-IDF project
- Enable PSRAM in `menuconfig`
+- Enable PSRAM in `menuconfig` (also set Flash and PSRAM frequiencies to 80MHz)
 - Include `esp_camera.h` in your code

 ### Using PlatformIO
@@ -75,17 +97,6 @@ However with a bit of patience and experimenting you'll figure the Kconfig out.

 If you miss-skip-ignore this critical step the camera module will compile but camera logic inside the library will be 'empty' because the Kconfig sets the proper #ifdef statements during the build process to initialize the selected cameras.  It's very not optional! 

-### Kconfig options
-
-| config                            | description                                                                                                                                                  | default                        |
-| --------------------------------- | ------------------------------------------------------------------------------------------------------------------------------------------------------------ | ------------------------------ |
-| CONFIG_OV2640_SUPPORT             | Support for OV2640 camera                                                                                                                                    | enabled                        |
-| CONFIG_OV7725_SUPPORT             | Support for OV7725 camera                                                                                                                                    | disabled                       |
-| CONFIG_OV3660_SUPPORT             | Support for OV3660 camera                                                                                                                                    | enabled                        |
-| CONFIG_OV5640_SUPPORT             | Support for OV5640 camera                                                                                                                                    | enabled                        |
-| CONFIG_SCCB_HARDWARE_I2C          | Enable this option if you want to use hardware I2C to control the camera. Disable this option to use software I2C.                                           | enabled                        |
-| CONFIG_SCCB_HARDWARE_I2C_PORT     | I2C peripheral to use for SCCB. Can be I2C0 and I2C1.                                                                                                        | CONFIG_SCCB_HARDWARE_I2C_PORT1 |
-| CONFIG_CAMERA_TASK_PINNED_TO_CORE | Pin the camera handle task to a certain core(0/1). It can also be done automatically choosing NO_AFFINITY. Can be CAMERA_CORE0, CAMERA_CORE1 or NO_AFFINITY. | CONFIG_CAMERA_CORE0            |

 ## Examples

@@ -132,8 +143,7 @@ static camera_config_t camera_config = {
    .pin_href = CAM_PIN_HREF,
    .pin_pclk = CAM_PIN_PCLK,

-    //XCLK 20MHz or 10MHz for OV2640 double FPS (Experimental)
-    .xclk_freq_hz = 20000000,
+    .xclk_freq_hz = 20000000,//EXPERIMENTAL: Set to 16MHz on ESP32-S2 or ESP32-S3 to enable EDMA mode
    .ledc_timer = LEDC_TIMER_0,
    .ledc_channel = LEDC_CHANNEL_0,

@@ -141,7 +151,8 @@ static camera_config_t camera_config = {
    .frame_size = FRAMESIZE_UXGA,//QQVGA-QXGA Do not use sizes above QVGA when not JPEG

    .jpeg_quality = 12, //0-63 lower number means higher quality
-    .fb_count = 1 //if more than one, i2s runs in continuous mode. Use only with JPEG
+    .fb_count = 1, //if more than one, i2s runs in continuous mode. Use only with JPEG
+    .grab_mode = CAMERA_GRAB_WHEN_EMPTY//CAMERA_GRAB_LATEST. Sets when buffers should be filled
 };

 esp_err_t camera_init(){
--- a/code/components/esp32-camera-master/component.mk
+++ b/code/components/esp32-camera-master/component.mk
@@ -1,4 +1,4 @@
 COMPONENT_ADD_INCLUDEDIRS := driver/include conversions/include
-COMPONENT_PRIV_INCLUDEDIRS := driver/private_include conversions/private_include sensors/private_include
-COMPONENT_SRCDIRS := driver conversions sensors
+COMPONENT_PRIV_INCLUDEDIRS := driver/private_include conversions/private_include sensors/private_include target/private_include
+COMPONENT_SRCDIRS := driver conversions sensors target target/esp32
 CXXFLAGS += -fno-rtti
--- a/code/components/esp32-camera-master/conversions/esp_jpg_decode.c
+++ b/code/components/esp32-camera-master/conversions/esp_jpg_decode.c
@@ -17,7 +17,11 @@
 #if ESP_IDF_VERSION_MAJOR >= 4 // IDF 4+
 #if CONFIG_IDF_TARGET_ESP32 // ESP32/PICO-D4
 #include "esp32/rom/tjpgd.h"
-#else 
+#elif CONFIG_IDF_TARGET_ESP32S2
+#include "tjpgd.h"
+#elif CONFIG_IDF_TARGET_ESP32S3
+#include "esp32s3/rom/tjpgd.h"
+#else
 #error Target CONFIG_IDF_TARGET is not supported
 #endif
 #else // ESP32 Before IDF 4.0
--- a/code/components/esp32-camera-master/conversions/include/img_converters.h
+++ b/code/components/esp32-camera-master/conversions/include/img_converters.h
@@ -22,6 +22,7 @@ extern "C" {
 #include <stdint.h>
 #include <stdbool.h>
 #include "esp_camera.h"
+#include "esp_jpg_decode.h"

 typedef size_t (* jpg_out_cb)(void * arg, size_t index, const void* data, size_t len);

@@ -120,6 +121,8 @@ bool frame2bmp(camera_fb_t * fb, uint8_t ** out, size_t * out_len);
 */
 bool fmt2rgb888(const uint8_t *src_buf, size_t src_len, pixformat_t format, uint8_t * rgb_buf);

+bool jpg2rgb565(const uint8_t *src, size_t src_len, uint8_t * out, jpg_scale_t scale);
+
 #ifdef __cplusplus
 }
 #endif
--- a/code/components/esp32-camera-master/conversions/to_bmp.c
+++ b/code/components/esp32-camera-master/conversions/to_bmp.c
@@ -24,6 +24,10 @@
 #if ESP_IDF_VERSION_MAJOR >= 4 // IDF 4+
 #if CONFIG_IDF_TARGET_ESP32 // ESP32/PICO-D4
 #include "esp32/spiram.h"
+#elif CONFIG_IDF_TARGET_ESP32S2
+#include "esp32s2/spiram.h"
+#elif CONFIG_IDF_TARGET_ESP32S3
+#include "esp32s3/spiram.h"
 #else 
 #error Target CONFIG_IDF_TARGET is not supported
 #endif
@@ -115,6 +119,54 @@ static bool _rgb_write(void * arg, uint16_t x, uint16_t y, uint16_t w, uint16_t
    return true;
 }

+static bool _rgb565_write(void * arg, uint16_t x, uint16_t y, uint16_t w, uint16_t h, uint8_t *data)
+{
+    rgb_jpg_decoder * jpeg = (rgb_jpg_decoder *)arg;
+    if(!data){
+        if(x == 0 && y == 0){
+            //write start
+            jpeg->width = w;
+            jpeg->height = h;
+            //if output is null, this is BMP
+            if(!jpeg->output){
+                jpeg->output = (uint8_t *)_malloc((w*h*3)+jpeg->data_offset);
+                if(!jpeg->output){
+                    return false;
+                }
+            }
+        } else {
+            //write end
+        }
+        return true;
+    }
+
+    size_t jw = jpeg->width*3;
+    size_t jw2 = jpeg->width*2;
+    size_t t = y * jw;
+    size_t t2 = y * jw2;
+    size_t b = t + (h * jw);
+    size_t l = x * 2;
+    uint8_t *out = jpeg->output+jpeg->data_offset;
+    uint8_t *o = out;
+    size_t iy, iy2, ix, ix2;
+
+    w = w * 3;
+
+    for(iy=t, iy2=t2; iy<b; iy+=jw, iy2+=jw2) {
+        o = out+iy2+l;
+        for(ix2=ix=0; ix<w; ix+= 3, ix2 +=2) {
+            uint16_t r = data[ix];
+            uint16_t g = data[ix+1];
+            uint16_t b = data[ix+2];
+            uint16_t c = ((r & 0xF8) << 8) | ((g & 0xFC) << 3) | (b >> 3);
+            o[ix2+1] = c>>8;
+            o[ix2] = c&0xff;
+        }
+        data+=w;
+    }
+    return true;
+}
+
 //input buffer
 static uint32_t _jpg_read(void * arg, size_t index, uint8_t *buf, size_t len)
 {
@@ -140,6 +192,21 @@ static bool jpg2rgb888(const uint8_t *src, size_t src_len, uint8_t * out, jpg_sc
    return true;
 }

+bool jpg2rgb565(const uint8_t *src, size_t src_len, uint8_t * out, jpg_scale_t scale)
+{
+    rgb_jpg_decoder jpeg;
+    jpeg.width = 0;
+    jpeg.height = 0;
+    jpeg.input = src;
+    jpeg.output = out;
+    jpeg.data_offset = 0;
+
+    if(esp_jpg_decode(src_len, scale, _jpg_read, _rgb565_write, (void*)&jpeg) != ESP_OK){
+        return false;
+    }
+    return true;
+}
+
 bool jpg2bmp(const uint8_t *src, size_t src_len, uint8_t ** out, size_t * out_len)
 {

@@ -317,7 +384,7 @@ bool fmt2bmp(uint8_t *src, size_t src_len, uint16_t width, uint16_t height, pixf
    }
    *out = out_buf;
    *out_len = out_size;
-    return true; 
+    return true;
 }

 bool frame2bmp(camera_fb_t * fb, uint8_t ** out, size_t * out_len)
--- a/code/components/esp32-camera-master/conversions/to_jpg.cpp
+++ b/code/components/esp32-camera-master/conversions/to_jpg.cpp
@@ -25,6 +25,10 @@
 #if ESP_IDF_VERSION_MAJOR >= 4 // IDF 4+
 #if CONFIG_IDF_TARGET_ESP32 // ESP32/PICO-D4
 #include "esp32/spiram.h"
+#elif CONFIG_IDF_TARGET_ESP32S2
+#include "esp32s2/spiram.h"
+#elif CONFIG_IDF_TARGET_ESP32S3
+#include "esp32s3/spiram.h"
 #else 
 #error Target CONFIG_IDF_TARGET is not supported
 #endif
@@ -195,7 +199,7 @@ public:
            return true;
        }
        if ((size_t)len > (max_len - index)) {
-            ESP_LOGW(TAG, "JPG output overflow: %d bytes", len - (max_len - index));
+            //ESP_LOGW(TAG, "JPG output overflow: %d bytes (%d,%d,%d)", len - (max_len - index), len, index, max_len);
            len = max_len - index;
        }
        if (len) {
@@ -215,7 +219,7 @@ bool fmt2jpg(uint8_t *src, size_t src_len, uint16_t width, uint16_t height, pixf
 {
    //todo: allocate proper buffer for holding JPEG data
    //this should be enough for CIF frame size
-    int jpg_buf_len = 64*1024;
+    int jpg_buf_len = 128*1024;


    uint8_t * jpg_buf = (uint8_t *)_malloc(jpg_buf_len);
--- a/code/components/esp32-camera-master/driver/cam_hal.c
+++ b/code/components/esp32-camera-master/driver/cam_hal.c
@@ -0,0 +1,483 @@
+// Copyright 2010-2020 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdio.h>
+#include <string.h>
+#include "esp_heap_caps.h"
+#include "ll_cam.h"
+#include "cam_hal.h"
+
+static const char *TAG = "cam_hal";
+
+static cam_obj_t *cam_obj = NULL;
+
+static const uint32_t JPEG_SOI_MARKER = 0xFFD8FF;  // written in little-endian for esp32
+static const uint16_t JPEG_EOI_MARKER = 0xD9FF;  // written in little-endian for esp32
+
+static int cam_verify_jpeg_soi(const uint8_t *inbuf, uint32_t length)
+{
+    uint32_t sig = *((uint32_t *)inbuf) & 0xFFFFFF;
+    if(sig != JPEG_SOI_MARKER) {
+        for (uint32_t i = 0; i < length; i++) {
+            sig = *((uint32_t *)(&inbuf[i])) & 0xFFFFFF;
+            if (sig == JPEG_SOI_MARKER) {
+                ESP_LOGW(TAG, "SOI: %d", i);
+                return i;
+            }
+        }
+        ESP_LOGW(TAG, "NO-SOI");
+        return -1;
+    }
+    return 0;
+}
+
+static int cam_verify_jpeg_eoi(const uint8_t *inbuf, uint32_t length)
+{
+    int offset = -1;
+    uint8_t *dptr = (uint8_t *)inbuf + length - 2;
+    while (dptr > inbuf) {
+        uint16_t sig = *((uint16_t *)dptr);
+        if (JPEG_EOI_MARKER == sig) {
+            offset = dptr - inbuf;
+            //ESP_LOGW(TAG, "EOI: %d", length - (offset + 2));
+            return offset;
+        }
+        dptr--;
+    }
+    return -1;
+}
+
+static bool cam_get_next_frame(int * frame_pos)
+{
+    if(!cam_obj->frames[*frame_pos].en){
+        for (int x = 0; x < cam_obj->frame_cnt; x++) {
+            if (cam_obj->frames[x].en) {
+                *frame_pos = x;
+                return true;
+            }
+        }
+    } else {
+        return true;
+    }
+    return false;
+}
+
+static bool cam_start_frame(int * frame_pos)
+{
+    if (cam_get_next_frame(frame_pos)) {
+        if(ll_cam_start(cam_obj, *frame_pos)){
+            // Vsync the frame manually
+            ll_cam_do_vsync(cam_obj);
+            uint64_t us = (uint64_t)esp_timer_get_time();
+            cam_obj->frames[*frame_pos].fb.timestamp.tv_sec = us / 1000000UL;
+            cam_obj->frames[*frame_pos].fb.timestamp.tv_usec = us % 1000000UL;
+            return true;
+        }
+    }
+    return false;
+}
+
+void IRAM_ATTR ll_cam_send_event(cam_obj_t *cam, cam_event_t cam_event, BaseType_t * HPTaskAwoken)
+{
+    if (xQueueSendFromISR(cam->event_queue, (void *)&cam_event, HPTaskAwoken) != pdTRUE) {
+        ll_cam_stop(cam);
+        cam->state = CAM_STATE_IDLE;
+        ESP_EARLY_LOGE(TAG, "EV-%s-OVF", cam_event==CAM_IN_SUC_EOF_EVENT ? "EOF" : "VSYNC");
+    }
+}
+
+//Copy fram from DMA dma_buffer to fram dma_buffer
+static void cam_task(void *arg)
+{
+    int cnt = 0;
+    int frame_pos = 0;
+    cam_obj->state = CAM_STATE_IDLE;
+    cam_event_t cam_event = 0;
+    
+    xQueueReset(cam_obj->event_queue);
+
+    while (1) {
+        xQueueReceive(cam_obj->event_queue, (void *)&cam_event, portMAX_DELAY);
+        DBG_PIN_SET(1);
+        switch (cam_obj->state) {
+
+            case CAM_STATE_IDLE: {
+                if (cam_event == CAM_VSYNC_EVENT) {
+                    //DBG_PIN_SET(1);
+                    if(cam_start_frame(&frame_pos)){
+                        cam_obj->frames[frame_pos].fb.len = 0;
+                        cam_obj->state = CAM_STATE_READ_BUF;
+                    }
+                    cnt = 0;
+                }
+            }
+            break;
+
+            case CAM_STATE_READ_BUF: {
+                camera_fb_t * frame_buffer_event = &cam_obj->frames[frame_pos].fb;
+                size_t pixels_per_dma = (cam_obj->dma_half_buffer_size * cam_obj->fb_bytes_per_pixel) / (cam_obj->dma_bytes_per_item * cam_obj->in_bytes_per_pixel);
+                
+                if (cam_event == CAM_IN_SUC_EOF_EVENT) {
+                    if(!cam_obj->psram_mode){
+                        if (cam_obj->fb_size < (frame_buffer_event->len + pixels_per_dma)) {
+                            ESP_LOGW(TAG, "FB-OVF");
+                            ll_cam_stop(cam_obj);
+                            DBG_PIN_SET(0);
+                            continue;
+                        }
+                        frame_buffer_event->len += ll_cam_memcpy(cam_obj,
+                            &frame_buffer_event->buf[frame_buffer_event->len], 
+                            &cam_obj->dma_buffer[(cnt % cam_obj->dma_half_buffer_cnt) * cam_obj->dma_half_buffer_size], 
+                            cam_obj->dma_half_buffer_size);
+                    }
+                    //Check for JPEG SOI in the first buffer. stop if not found
+                    if (cam_obj->jpeg_mode && cnt == 0 && cam_verify_jpeg_soi(frame_buffer_event->buf, frame_buffer_event->len) != 0) {
+                        ll_cam_stop(cam_obj);
+                        cam_obj->state = CAM_STATE_IDLE;
+                    }
+                    cnt++;
+
+                } else if (cam_event == CAM_VSYNC_EVENT) {
+                    //DBG_PIN_SET(1);
+                    ll_cam_stop(cam_obj);
+
+                    if (cnt || !cam_obj->jpeg_mode || cam_obj->psram_mode) {
+                        if (cam_obj->jpeg_mode) {
+                            if (!cam_obj->psram_mode) {
+                                if (cam_obj->fb_size < (frame_buffer_event->len + pixels_per_dma)) {
+                                    ESP_LOGW(TAG, "FB-OVF");
+                                    cnt--;
+                                } else {
+                                    frame_buffer_event->len += ll_cam_memcpy(cam_obj,
+                                        &frame_buffer_event->buf[frame_buffer_event->len], 
+                                        &cam_obj->dma_buffer[(cnt % cam_obj->dma_half_buffer_cnt) * cam_obj->dma_half_buffer_size], 
+                                        cam_obj->dma_half_buffer_size);
+                                }
+                            }
+                            cnt++;
+                        }
+
+                        cam_obj->frames[frame_pos].en = 0;
+
+                        if (cam_obj->psram_mode) {
+                            if (cam_obj->jpeg_mode) {
+                                frame_buffer_event->len = cnt * cam_obj->dma_half_buffer_size;
+                            } else {
+                                frame_buffer_event->len = cam_obj->recv_size;
+                            }
+                        } else if (!cam_obj->jpeg_mode) {
+                            if (frame_buffer_event->len != cam_obj->fb_size) {
+                                cam_obj->frames[frame_pos].en = 1;
+                                ESP_LOGE(TAG, "FB-SIZE: %u != %u", frame_buffer_event->len, cam_obj->fb_size);
+                            }
+                        }
+                        //send frame
+                        if(!cam_obj->frames[frame_pos].en && xQueueSend(cam_obj->frame_buffer_queue, (void *)&frame_buffer_event, 0) != pdTRUE) {
+                            //pop frame buffer from the queue
+                            camera_fb_t * fb2 = NULL;
+                            if(xQueueReceive(cam_obj->frame_buffer_queue, &fb2, 0) == pdTRUE) {
+                                //push the new frame to the end of the queue
+                                if (xQueueSend(cam_obj->frame_buffer_queue, (void *)&frame_buffer_event, 0) != pdTRUE) {
+                                    cam_obj->frames[frame_pos].en = 1;
+                                    ESP_LOGE(TAG, "FBQ-SND");
+                                }
+                                //free the popped buffer
+                                cam_give(fb2);
+                            } else {
+                                //queue is full and we could not pop a frame from it
+                                cam_obj->frames[frame_pos].en = 1;
+                                ESP_LOGE(TAG, "FBQ-RCV");
+                            }
+                        }
+                    }
+
+                    if(!cam_start_frame(&frame_pos)){
+                        cam_obj->state = CAM_STATE_IDLE;
+                    } else {
+                        cam_obj->frames[frame_pos].fb.len = 0;
+                    }
+                    cnt = 0;
+                }
+            }
+            break;
+        }
+        DBG_PIN_SET(0);
+    }
+}
+
+static lldesc_t * allocate_dma_descriptors(uint32_t count, uint16_t size, uint8_t * buffer)
+{
+    lldesc_t *dma = (lldesc_t *)heap_caps_malloc(count * sizeof(lldesc_t), MALLOC_CAP_DMA);
+    if (dma == NULL) {
+        return dma;
+    }
+
+    for (int x = 0; x < count; x++) {
+        dma[x].size = size;
+        dma[x].length = 0;
+        dma[x].sosf = 0;
+        dma[x].eof = 0;
+        dma[x].owner = 1;
+        dma[x].buf = (buffer + size * x);
+        dma[x].empty = (uint32_t)&dma[(x + 1) % count];
+    }
+    return dma;
+}
+
+static esp_err_t cam_dma_config(const camera_config_t *config)
+{
+    bool ret = ll_cam_dma_sizes(cam_obj);
+    if (0 == ret) {
+        return ESP_FAIL;
+    }
+
+    cam_obj->dma_node_cnt = (cam_obj->dma_buffer_size) / cam_obj->dma_node_buffer_size; // Number of DMA nodes
+    cam_obj->frame_copy_cnt = cam_obj->recv_size / cam_obj->dma_half_buffer_size; // Number of interrupted copies, ping-pong copy
+
+    ESP_LOGI(TAG, "buffer_size: %d, half_buffer_size: %d, node_buffer_size: %d, node_cnt: %d, total_cnt: %d", 
+             cam_obj->dma_buffer_size, cam_obj->dma_half_buffer_size, cam_obj->dma_node_buffer_size, cam_obj->dma_node_cnt, cam_obj->frame_copy_cnt);
+
+    cam_obj->dma_buffer = NULL;
+    cam_obj->dma = NULL;
+
+    cam_obj->frames = (cam_frame_t *)heap_caps_calloc(1, cam_obj->frame_cnt * sizeof(cam_frame_t), MALLOC_CAP_DEFAULT);
+    CAM_CHECK(cam_obj->frames != NULL, "frames malloc failed", ESP_FAIL);
+
+    uint8_t dma_align = 0;
+    size_t fb_size = cam_obj->fb_size;
+    if (cam_obj->psram_mode) {
+        dma_align = ll_cam_get_dma_align(cam_obj);
+        if (cam_obj->fb_size < cam_obj->recv_size) {
+            fb_size = cam_obj->recv_size;
+        }
+    }
+
+    /* Allocate memory for frame buffer */
+    size_t alloc_size = fb_size * sizeof(uint8_t) + dma_align;
+    uint32_t _caps = MALLOC_CAP_8BIT;
+    if (CAMERA_FB_IN_DRAM == config->fb_location) {
+        _caps |= MALLOC_CAP_INTERNAL;
+    } else {
+        _caps |= MALLOC_CAP_SPIRAM;
+    }
+    for (int x = 0; x < cam_obj->frame_cnt; x++) {
+        cam_obj->frames[x].dma = NULL;
+        cam_obj->frames[x].fb_offset = 0;
+        cam_obj->frames[x].en = 0;
+        ESP_LOGI(TAG, "Allocating %d Byte frame buffer in %s", alloc_size, _caps & MALLOC_CAP_SPIRAM ? "PSRAM" : "OnBoard RAM");
+        cam_obj->frames[x].fb.buf = (uint8_t *)heap_caps_malloc(alloc_size, _caps);
+        CAM_CHECK(cam_obj->frames[x].fb.buf != NULL, "frame buffer malloc failed", ESP_FAIL);
+        if (cam_obj->psram_mode) {
+            //align PSRAM buffer. TODO: save the offset so proper address can be freed later
+            cam_obj->frames[x].fb_offset = dma_align - ((uint32_t)cam_obj->frames[x].fb.buf & (dma_align - 1));
+            cam_obj->frames[x].fb.buf += cam_obj->frames[x].fb_offset;
+            ESP_LOGI(TAG, "Frame[%d]: Offset: %u, Addr: 0x%08X", x, cam_obj->frames[x].fb_offset, (uint32_t)cam_obj->frames[x].fb.buf);
+            cam_obj->frames[x].dma = allocate_dma_descriptors(cam_obj->dma_node_cnt, cam_obj->dma_node_buffer_size, cam_obj->frames[x].fb.buf);
+            CAM_CHECK(cam_obj->frames[x].dma != NULL, "frame dma malloc failed", ESP_FAIL);
+        }
+        cam_obj->frames[x].en = 1;
+    }
+
+    if (!cam_obj->psram_mode) {
+        cam_obj->dma_buffer = (uint8_t *)heap_caps_malloc(cam_obj->dma_buffer_size * sizeof(uint8_t), MALLOC_CAP_DMA);
+        if(NULL == cam_obj->dma_buffer) {
+            ESP_LOGE(TAG,"%s(%d): DMA buffer %d Byte malloc failed, the current largest free block:%d Byte", __FUNCTION__, __LINE__, 
+                     cam_obj->dma_buffer_size, heap_caps_get_largest_free_block(MALLOC_CAP_DMA));
+            return ESP_FAIL;
+        }
+
+        cam_obj->dma = allocate_dma_descriptors(cam_obj->dma_node_cnt, cam_obj->dma_node_buffer_size, cam_obj->dma_buffer);
+        CAM_CHECK(cam_obj->dma != NULL, "dma malloc failed", ESP_FAIL);
+    }
+
+    return ESP_OK;
+}
+
+esp_err_t cam_init(const camera_config_t *config)
+{
+    CAM_CHECK(NULL != config, "config pointer is invalid", ESP_ERR_INVALID_ARG);
+
+    esp_err_t ret = ESP_OK;
+    cam_obj = (cam_obj_t *)heap_caps_calloc(1, sizeof(cam_obj_t), MALLOC_CAP_DMA);
+    CAM_CHECK(NULL != cam_obj, "lcd_cam object malloc error", ESP_ERR_NO_MEM);
+
+    cam_obj->swap_data = 0;
+    cam_obj->vsync_pin = config->pin_vsync;
+    cam_obj->vsync_invert = true;
+
+    ll_cam_set_pin(cam_obj, config);
+    ret = ll_cam_config(cam_obj, config);
+    CAM_CHECK_GOTO(ret == ESP_OK, "ll_cam initialize failed", err);
+
+#if CAMERA_DBG_PIN_ENABLE
+    PIN_FUNC_SELECT(GPIO_PIN_MUX_REG[DBG_PIN_NUM], PIN_FUNC_GPIO);
+    gpio_set_direction(DBG_PIN_NUM, GPIO_MODE_OUTPUT);
+    gpio_set_pull_mode(DBG_PIN_NUM, GPIO_FLOATING);
+#endif
+
+    ESP_LOGI(TAG, "cam init ok");
+    return ESP_OK;
+
+err:
+    free(cam_obj);
+    cam_obj = NULL;
+    return ESP_FAIL;
+}
+
+esp_err_t cam_config(const camera_config_t *config, framesize_t frame_size, uint16_t sensor_pid)
+{
+    CAM_CHECK(NULL != config, "config pointer is invalid", ESP_ERR_INVALID_ARG);
+    esp_err_t ret = ESP_OK;
+
+    ret = ll_cam_set_sample_mode(cam_obj, (pixformat_t)config->pixel_format, config->xclk_freq_hz, sensor_pid);
+
+    cam_obj->jpeg_mode = config->pixel_format == PIXFORMAT_JPEG;
+#if CONFIG_IDF_TARGET_ESP32
+    cam_obj->psram_mode = false;
+#else
+    cam_obj->psram_mode = (config->xclk_freq_hz == 16000000);
+#endif
+    cam_obj->frame_cnt = config->fb_count;
+    cam_obj->width = resolution[frame_size].width;
+    cam_obj->height = resolution[frame_size].height;
+
+    if(cam_obj->jpeg_mode){
+        cam_obj->recv_size = cam_obj->width * cam_obj->height / 5;
+        cam_obj->fb_size = cam_obj->recv_size;
+    } else {
+        cam_obj->recv_size = cam_obj->width * cam_obj->height * cam_obj->in_bytes_per_pixel;
+        cam_obj->fb_size = cam_obj->width * cam_obj->height * cam_obj->fb_bytes_per_pixel;
+    }
+    
+    ret = cam_dma_config(config);
+    CAM_CHECK_GOTO(ret == ESP_OK, "cam_dma_config failed", err);
+
+    cam_obj->event_queue = xQueueCreate(cam_obj->dma_half_buffer_cnt - 1, sizeof(cam_event_t));
+    CAM_CHECK_GOTO(cam_obj->event_queue != NULL, "event_queue create failed", err);
+
+    size_t frame_buffer_queue_len = cam_obj->frame_cnt;
+    if (config->grab_mode == CAMERA_GRAB_LATEST && cam_obj->frame_cnt > 1) {
+        frame_buffer_queue_len = cam_obj->frame_cnt - 1;
+    }
+    cam_obj->frame_buffer_queue = xQueueCreate(frame_buffer_queue_len, sizeof(camera_fb_t*));
+    CAM_CHECK_GOTO(cam_obj->frame_buffer_queue != NULL, "frame_buffer_queue create failed", err);
+
+    ret = ll_cam_init_isr(cam_obj);
+    CAM_CHECK_GOTO(ret == ESP_OK, "cam intr alloc failed", err);
+
+    
+#if CONFIG_CAMERA_CORE0
+    xTaskCreatePinnedToCore(cam_task, "cam_task", 2048, NULL, configMAX_PRIORITIES - 2, &cam_obj->task_handle, 0);
+#elif CONFIG_CAMERA_CORE1
+    xTaskCreatePinnedToCore(cam_task, "cam_task", 2048, NULL, configMAX_PRIORITIES - 2, &cam_obj->task_handle, 1);
+#else
+    xTaskCreate(cam_task, "cam_task", 2048, NULL, configMAX_PRIORITIES - 2, &cam_obj->task_handle);
+#endif
+
+    ESP_LOGI(TAG, "cam config ok");
+    return ESP_OK;
+
+err:
+    cam_deinit();
+    return ESP_FAIL;
+}
+
+esp_err_t cam_deinit(void)
+{
+    if (!cam_obj) {
+        return ESP_FAIL;
+    }
+
+    cam_stop();
+    if (cam_obj->task_handle) {
+        vTaskDelete(cam_obj->task_handle);
+    }
+    if (cam_obj->event_queue) {
+        vQueueDelete(cam_obj->event_queue);
+    }
+    if (cam_obj->frame_buffer_queue) {
+        vQueueDelete(cam_obj->frame_buffer_queue);
+    }
+    if (cam_obj->dma) {
+        free(cam_obj->dma);
+    }
+    if (cam_obj->dma_buffer) {
+        free(cam_obj->dma_buffer);
+    }
+    if (cam_obj->frames) {
+        for (int x = 0; x < cam_obj->frame_cnt; x++) {
+            free(cam_obj->frames[x].fb.buf - cam_obj->frames[x].fb_offset);
+            if (cam_obj->frames[x].dma) {
+                free(cam_obj->frames[x].dma);
+            }
+        }
+        free(cam_obj->frames);
+    }
+
+    ll_cam_deinit(cam_obj);
+
+    free(cam_obj);
+    cam_obj = NULL;
+    return ESP_OK;
+}
+
+void cam_stop(void)
+{
+    ll_cam_vsync_intr_enable(cam_obj, false);
+    ll_cam_stop(cam_obj);
+}
+
+void cam_start(void)
+{
+    ll_cam_vsync_intr_enable(cam_obj, true);
+}
+
+camera_fb_t *cam_take(TickType_t timeout)
+{
+    camera_fb_t *dma_buffer = NULL;
+    TickType_t start = xTaskGetTickCount();
+    xQueueReceive(cam_obj->frame_buffer_queue, (void *)&dma_buffer, timeout);
+    if (dma_buffer) {
+        if(cam_obj->jpeg_mode){
+            // find the end marker for JPEG. Data after that can be discarded
+            int offset_e = cam_verify_jpeg_eoi(dma_buffer->buf, dma_buffer->len);
+            if (offset_e >= 0) {
+                // adjust buffer length
+                dma_buffer->len = offset_e + sizeof(JPEG_EOI_MARKER);
+                return dma_buffer;
+            } else {
+                ESP_LOGW(TAG, "NO-EOI");
+                cam_give(dma_buffer);
+                return cam_take(timeout - (xTaskGetTickCount() - start));//recurse!!!!
+            }
+        } else if(cam_obj->psram_mode && cam_obj->in_bytes_per_pixel != cam_obj->fb_bytes_per_pixel){
+            //currently this is used only for YUV to GRAYSCALE
+            dma_buffer->len = ll_cam_memcpy(cam_obj, dma_buffer->buf, dma_buffer->buf, dma_buffer->len);
+        }
+        return dma_buffer;
+    } else {
+        ESP_LOGW(TAG, "Failed to get the frame on time!");
+    }
+    return NULL;
+}
+
+void cam_give(camera_fb_t *dma_buffer)
+{
+    for (int x = 0; x < cam_obj->frame_cnt; x++) {
+        if (&cam_obj->frames[x].fb == dma_buffer) {
+            cam_obj->frames[x].en = 1;
+            break;
+        }
+    }
+}
--- a/code/components/esp32-camera-master/driver/camera.c
+++ b/code/components/esp32-camera-master/driver/camera.c
--- a/code/components/esp32-camera-master/driver/esp_camera.c
+++ b/code/components/esp32-camera-master/driver/esp_camera.c
@@ -0,0 +1,421 @@
+// Copyright 2015-2016 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include "time.h"
+#include "sys/time.h"
+#include "freertos/FreeRTOS.h"
+#include "freertos/task.h"
+#include "driver/gpio.h"
+#include "esp_system.h"
+#include "nvs_flash.h"
+#include "nvs.h"
+#include "sensor.h"
+#include "sccb.h"
+#include "cam_hal.h"
+#include "esp_camera.h"
+#include "xclk.h"
+#if CONFIG_OV2640_SUPPORT
+#include "ov2640.h"
+#endif
+#if CONFIG_OV7725_SUPPORT
+#include "ov7725.h"
+#endif
+#if CONFIG_OV3660_SUPPORT
+#include "ov3660.h"
+#endif
+#if CONFIG_OV5640_SUPPORT
+#include "ov5640.h"
+#endif
+#if CONFIG_NT99141_SUPPORT
+#include "nt99141.h"
+#endif
+#if CONFIG_OV7670_SUPPORT
+#include "ov7670.h"
+#endif
+#if CONFIG_GC2145_SUPPORT
+#include "gc2145.h"
+#endif
+#if CONFIG_GC032A_SUPPORT
+#include "gc032a.h"
+#endif
+#if CONFIG_GC0308_SUPPORT
+#include "gc0308.h"
+#endif
+#if CONFIG_BF3005_SUPPORT
+#include "bf3005.h"
+#endif
+
+#if defined(ARDUINO_ARCH_ESP32) && defined(CONFIG_ARDUHAL_ESP_LOG)
+#include "esp32-hal-log.h"
+#define TAG ""
+#else
+#include "esp_log.h"
+static const char *TAG = "camera";
+#endif
+
+typedef struct {
+    sensor_t sensor;
+    camera_fb_t fb;
+} camera_state_t;
+
+static const char *CAMERA_SENSOR_NVS_KEY = "sensor";
+static const char *CAMERA_PIXFORMAT_NVS_KEY = "pixformat";
+static camera_state_t *s_state = NULL;
+
+#if CONFIG_IDF_TARGET_ESP32S3 // LCD_CAM module of ESP32-S3 will generate xclk
+#define CAMERA_ENABLE_OUT_CLOCK(v)
+#define CAMERA_DISABLE_OUT_CLOCK()
+#else
+#define CAMERA_ENABLE_OUT_CLOCK(v) camera_enable_out_clock((v))
+#define CAMERA_DISABLE_OUT_CLOCK() camera_disable_out_clock()
+#endif
+
+typedef struct {
+    int (*detect)(int slv_addr, sensor_id_t *id);
+    int (*init)(sensor_t *sensor);
+} sensor_func_t;
+
+static const sensor_func_t g_sensors[] = {
+#if CONFIG_OV7725_SUPPORT
+    {ov7725_detect, ov7725_init},
+#endif
+#if CONFIG_OV7670_SUPPORT
+    {ov7670_detect, ov7670_init},
+#endif
+#if CONFIG_OV2640_SUPPORT
+    {ov2640_detect, ov2640_init},
+#endif
+#if CONFIG_OV3660_SUPPORT
+    {ov3660_detect, ov3660_init},
+#endif
+#if CONFIG_OV5640_SUPPORT
+    {ov5640_detect, ov5640_init},
+#endif
+#if CONFIG_NT99141_SUPPORT
+    {nt99141_detect, nt99141_init},
+#endif
+#if CONFIG_GC2145_SUPPORT
+    {gc2145_detect, gc2145_init},
+#endif
+#if CONFIG_GC032A_SUPPORT
+    {gc032a_detect, gc032a_init},
+#endif
+#if CONFIG_GC0308_SUPPORT
+    {gc0308_detect, gc0308_init},
+#endif
+#if CONFIG_BF3005_SUPPORT
+    {bf3005_detect, bf3005_init},
+#endif
+};
+
+static esp_err_t camera_probe(const camera_config_t *config, camera_model_t *out_camera_model)
+{
+    *out_camera_model = CAMERA_NONE;
+    if (s_state != NULL) {
+        return ESP_ERR_INVALID_STATE;
+    }
+
+    s_state = (camera_state_t *) calloc(sizeof(camera_state_t), 1);
+    if (!s_state) {
+        return ESP_ERR_NO_MEM;
+    }
+
+    if (config->pin_xclk >= 0) {
+        ESP_LOGD(TAG, "Enabling XCLK output");
+        CAMERA_ENABLE_OUT_CLOCK(config);
+    }
+
+    if (config->pin_sscb_sda != -1) {
+        ESP_LOGD(TAG, "Initializing SSCB");
+        SCCB_Init(config->pin_sscb_sda, config->pin_sscb_scl);
+    }
+
+    if (config->pin_pwdn >= 0) {
+        ESP_LOGD(TAG, "Resetting camera by power down line");
+        gpio_config_t conf = { 0 };
+        conf.pin_bit_mask = 1LL << config->pin_pwdn;
+        conf.mode = GPIO_MODE_OUTPUT;
+        gpio_config(&conf);
+
+        // carefull, logic is inverted compared to reset pin
+        gpio_set_level(config->pin_pwdn, 1);
+        vTaskDelay(10 / portTICK_PERIOD_MS);
+        gpio_set_level(config->pin_pwdn, 0);
+        vTaskDelay(10 / portTICK_PERIOD_MS);
+    }
+
+    if (config->pin_reset >= 0) {
+        ESP_LOGD(TAG, "Resetting camera");
+        gpio_config_t conf = { 0 };
+        conf.pin_bit_mask = 1LL << config->pin_reset;
+        conf.mode = GPIO_MODE_OUTPUT;
+        gpio_config(&conf);
+
+        gpio_set_level(config->pin_reset, 0);
+        vTaskDelay(10 / portTICK_PERIOD_MS);
+        gpio_set_level(config->pin_reset, 1);
+        vTaskDelay(10 / portTICK_PERIOD_MS);
+    }
+
+
+    ESP_LOGD(TAG, "Searching for camera address");
+    vTaskDelay(10 / portTICK_PERIOD_MS);
+
+    uint8_t slv_addr = SCCB_Probe();
+
+    if (slv_addr == 0) {
+        CAMERA_DISABLE_OUT_CLOCK();
+        return ESP_ERR_NOT_FOUND;
+    }
+
+    ESP_LOGI(TAG, "Detected camera at address=0x%02x", slv_addr);
+    s_state->sensor.slv_addr = slv_addr;
+    s_state->sensor.xclk_freq_hz = config->xclk_freq_hz;
+
+    /**
+     * Read sensor ID and then initialize sensor
+     * Attention: Some sensors have the same SCCB address. Therefore, several attempts may be made in the detection process
+     */
+    sensor_id_t *id = &s_state->sensor.id;
+    for (size_t i = 0; i < sizeof(g_sensors) / sizeof(sensor_func_t); i++) {
+        if (g_sensors[i].detect(slv_addr, id)) {
+            camera_sensor_info_t *info = esp_camera_sensor_get_info(id);
+            if (NULL != info) {
+                *out_camera_model = info->model;
+                ESP_LOGI(TAG, "Detected %s camera", info->name);
+                g_sensors[i].init(&s_state->sensor);
+                break;
+            }
+        }
+    }
+
+    if (CAMERA_NONE == *out_camera_model) { //If no supported sensors are detected
+        CAMERA_DISABLE_OUT_CLOCK();
+        ESP_LOGE(TAG, "Detected camera not supported.");
+        return ESP_ERR_NOT_SUPPORTED;
+    }
+
+    ESP_LOGI(TAG, "Camera PID=0x%02x VER=0x%02x MIDL=0x%02x MIDH=0x%02x",
+             id->PID, id->VER, id->MIDH, id->MIDL);
+
+    ESP_LOGD(TAG, "Doing SW reset of sensor");
+    vTaskDelay(10 / portTICK_PERIOD_MS);
+    s_state->sensor.reset(&s_state->sensor);
+
+    return ESP_OK;
+}
+
+esp_err_t esp_camera_init(const camera_config_t *config)
+{
+    esp_err_t err;
+    err = cam_init(config);
+    if (err != ESP_OK) {
+        ESP_LOGE(TAG, "Camera init failed with error 0x%x", err);
+        return err;
+    }
+
+    camera_model_t camera_model = CAMERA_NONE;
+    err = camera_probe(config, &camera_model);
+    if (err != ESP_OK) {
+        ESP_LOGE(TAG, "Camera probe failed with error 0x%x(%s)", err, esp_err_to_name(err));
+        goto fail;
+    }
+
+    framesize_t frame_size = (framesize_t) config->frame_size;
+    pixformat_t pix_format = (pixformat_t) config->pixel_format;
+
+    if (PIXFORMAT_JPEG == pix_format && (!camera_sensor[camera_model].support_jpeg)) {
+        ESP_LOGE(TAG, "JPEG format is not supported on this sensor");
+        err = ESP_ERR_NOT_SUPPORTED;
+        goto fail;
+    }
+
+    if (frame_size > camera_sensor[camera_model].max_size) {
+        ESP_LOGW(TAG, "The frame size exceeds the maximum for this sensor, it will be forced to the maximum possible value");
+        frame_size = camera_sensor[camera_model].max_size;
+    }
+
+    err = cam_config(config, frame_size, s_state->sensor.id.PID);
+    if (err != ESP_OK) {
+        ESP_LOGE(TAG, "Camera config failed with error 0x%x", err);
+        goto fail;
+    }
+
+    s_state->sensor.status.framesize = frame_size;
+    s_state->sensor.pixformat = pix_format;
+    ESP_LOGD(TAG, "Setting frame size to %dx%d", resolution[frame_size].width, resolution[frame_size].height);
+    if (s_state->sensor.set_framesize(&s_state->sensor, frame_size) != 0) {
+        ESP_LOGE(TAG, "Failed to set frame size");
+        err = ESP_ERR_CAMERA_FAILED_TO_SET_FRAME_SIZE;
+        goto fail;
+    }
+    s_state->sensor.set_pixformat(&s_state->sensor, pix_format);
+
+    if (s_state->sensor.id.PID == OV2640_PID) {
+        s_state->sensor.set_gainceiling(&s_state->sensor, GAINCEILING_2X);
+        s_state->sensor.set_bpc(&s_state->sensor, false);
+        s_state->sensor.set_wpc(&s_state->sensor, true);
+        s_state->sensor.set_lenc(&s_state->sensor, true);
+    }
+
+    if (pix_format == PIXFORMAT_JPEG) {
+        s_state->sensor.set_quality(&s_state->sensor, config->jpeg_quality);
+    }
+    s_state->sensor.init_status(&s_state->sensor);
+
+    cam_start();
+
+    return ESP_OK;
+
+fail:
+    esp_camera_deinit();
+    return err;
+}
+
+esp_err_t esp_camera_deinit()
+{
+    esp_err_t ret = cam_deinit();
+    CAMERA_DISABLE_OUT_CLOCK();
+    if (s_state) {
+        SCCB_Deinit();
+
+        free(s_state);
+        s_state = NULL;
+    }
+
+    return ret;
+}
+
+#define FB_GET_TIMEOUT (4000 / portTICK_PERIOD_MS)
+
+camera_fb_t *esp_camera_fb_get()
+{
+    if (s_state == NULL) {
+        return NULL;
+    }
+    camera_fb_t *fb = cam_take(FB_GET_TIMEOUT);
+    //set the frame properties
+    if (fb) {
+        fb->width = resolution[s_state->sensor.status.framesize].width;
+        fb->height = resolution[s_state->sensor.status.framesize].height;
+        fb->format = s_state->sensor.pixformat;
+    }
+    return fb;
+}
+
+void esp_camera_fb_return(camera_fb_t *fb)
+{
+    if (s_state == NULL) {
+        return;
+    }
+    cam_give(fb);
+}
+
+sensor_t *esp_camera_sensor_get()
+{
+    if (s_state == NULL) {
+        return NULL;
+    }
+    return &s_state->sensor;
+}
+
+esp_err_t esp_camera_save_to_nvs(const char *key)
+{
+#if ESP_IDF_VERSION_MAJOR > 3
+    nvs_handle_t handle;
+#else
+    nvs_handle handle;
+#endif
+    esp_err_t ret = nvs_open(key, NVS_READWRITE, &handle);
+
+    if (ret == ESP_OK) {
+        sensor_t *s = esp_camera_sensor_get();
+        if (s != NULL) {
+            ret = nvs_set_blob(handle, CAMERA_SENSOR_NVS_KEY, &s->status, sizeof(camera_status_t));
+            if (ret == ESP_OK) {
+                uint8_t pf = s->pixformat;
+                ret = nvs_set_u8(handle, CAMERA_PIXFORMAT_NVS_KEY, pf);
+            }
+            return ret;
+        } else {
+            return ESP_ERR_CAMERA_NOT_DETECTED;
+        }
+        nvs_close(handle);
+        return ret;
+    } else {
+        return ret;
+    }
+}
+
+esp_err_t esp_camera_load_from_nvs(const char *key)
+{
+#if ESP_IDF_VERSION_MAJOR > 3
+    nvs_handle_t handle;
+#else
+    nvs_handle handle;
+#endif
+    uint8_t pf;
+
+    esp_err_t ret = nvs_open(key, NVS_READWRITE, &handle);
+
+    if (ret == ESP_OK) {
+        sensor_t *s = esp_camera_sensor_get();
+        camera_status_t st;
+        if (s != NULL) {
+            size_t size = sizeof(camera_status_t);
+            ret = nvs_get_blob(handle, CAMERA_SENSOR_NVS_KEY, &st, &size);
+            if (ret == ESP_OK) {
+                s->set_ae_level(s, st.ae_level);
+                s->set_aec2(s, st.aec2);
+                s->set_aec_value(s, st.aec_value);
+                s->set_agc_gain(s, st.agc_gain);
+                s->set_awb_gain(s, st.awb_gain);
+                s->set_bpc(s, st.bpc);
+                s->set_brightness(s, st.brightness);
+                s->set_colorbar(s, st.colorbar);
+                s->set_contrast(s, st.contrast);
+                s->set_dcw(s, st.dcw);
+                s->set_denoise(s, st.denoise);
+                s->set_exposure_ctrl(s, st.aec);
+                s->set_framesize(s, st.framesize);
+                s->set_gain_ctrl(s, st.agc);
+                s->set_gainceiling(s, st.gainceiling);
+                s->set_hmirror(s, st.hmirror);
+                s->set_lenc(s, st.lenc);
+                s->set_quality(s, st.quality);
+                s->set_raw_gma(s, st.raw_gma);
+                s->set_saturation(s, st.saturation);
+                s->set_sharpness(s, st.sharpness);
+                s->set_special_effect(s, st.special_effect);
+                s->set_vflip(s, st.vflip);
+                s->set_wb_mode(s, st.wb_mode);
+                s->set_whitebal(s, st.awb);
+                s->set_wpc(s, st.wpc);
+            }
+            ret = nvs_get_u8(handle, CAMERA_PIXFORMAT_NVS_KEY, &pf);
+            if (ret == ESP_OK) {
+                s->set_pixformat(s, pf);
+            }
+        } else {
+            return ESP_ERR_CAMERA_NOT_DETECTED;
+        }
+        nvs_close(handle);
+        return ret;
+    } else {
+        ESP_LOGW(TAG, "Error (%d) opening nvs key \"%s\"", ret, key);
+        return ret;
+    }
+}
--- a/code/components/esp32-camera-master/driver/include/esp_camera.h
+++ b/code/components/esp32-camera-master/driver/include/esp_camera.h
@@ -38,7 +38,8 @@
        .pixel_format   = PIXFORMAT_JPEG,
        .frame_size     = FRAMESIZE_SVGA,
        .jpeg_quality   = 10,
-        .fb_count       = 2
+        .fb_count       = 2,
+        .grab_mode      = CAMERA_GRAB_WHEN_EMPTY
    };

    esp_err_t camera_example_init(){
@@ -74,6 +75,22 @@
 extern "C" {
 #endif

+/**
+ * @brief Configuration structure for camera initialization
+ */
+typedef enum {
+    CAMERA_GRAB_WHEN_EMPTY,         /*!< Fills buffers when they are empty. Less resources but first 'fb_count' frames might be old */
+    CAMERA_GRAB_LATEST              /*!< Except when 1 frame buffer is used, queue will always contain the last 'fb_count' frames */
+} camera_grab_mode_t;
+
+/**
+ * @brief Camera frame buffer location 
+ */
+typedef enum {
+    CAMERA_FB_IN_PSRAM,         /*!< Frame buffer is placed in external PSRAM */
+    CAMERA_FB_IN_DRAM           /*!< Frame buffer is placed in internal DRAM */
+} camera_fb_location_t;
+
 /**
 * @brief Configuration structure for camera initialization
 */
@@ -95,7 +112,7 @@ typedef struct {
    int pin_href;                   /*!< GPIO pin for camera HREF line */
    int pin_pclk;                   /*!< GPIO pin for camera PCLK line */

-    int xclk_freq_hz;               /*!< Frequency of XCLK signal, in Hz. Either 20KHz or 10KHz for OV2640 double FPS (Experimental) */
+    int xclk_freq_hz;               /*!< Frequency of XCLK signal, in Hz. EXPERIMENTAL: Set to 16MHz on ESP32-S2 or ESP32-S3 to enable EDMA mode */

    ledc_timer_t ledc_timer;        /*!< LEDC timer to be used for generating XCLK  */
    ledc_channel_t ledc_channel;    /*!< LEDC channel to be used for generating XCLK  */
@@ -105,6 +122,8 @@ typedef struct {

    int jpeg_quality;               /*!< Quality of JPEG output. 0-63 lower means higher quality  */
    size_t fb_count;                /*!< Number of frame buffers to be allocated. If more than one, then each frame will be acquired (double speed)  */
+    camera_fb_location_t fb_location; /*!< The location where the frame buffer will be allocated */
+    camera_grab_mode_t grab_mode;   /*!< When buffers should be filled */
 } camera_config_t;

 /**
--- a/code/components/esp32-camera-master/driver/include/sensor.h
+++ b/code/components/esp32-camera-master/driver/include/sensor.h
@@ -11,13 +11,51 @@
 #include <stdint.h>
 #include <stdbool.h>

-#define NT99141_PID     (0x14)
-#define OV9650_PID     (0x96)
-#define OV7725_PID     (0x77)
-#define OV2640_PID     (0x26)
-#define OV3660_PID     (0x36)
-#define OV5640_PID     (0x56)
-#define OV7670_PID     (0x76)
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+typedef enum {
+    OV9650_PID = 0x96,
+    OV7725_PID = 0x77,
+    OV2640_PID = 0x26,
+    OV3660_PID = 0x3660,
+    OV5640_PID = 0x5640,
+    OV7670_PID = 0x76,
+    NT99141_PID = 0x1410,
+    GC2145_PID = 0x2145,
+    GC032A_PID = 0x232a,
+    GC0308_PID = 0x9b,
+    BF3005_PID = 0x30,
+} camera_pid_t;
+
+typedef enum {
+    CAMERA_OV7725,
+    CAMERA_OV2640,
+    CAMERA_OV3660,
+    CAMERA_OV5640,
+    CAMERA_OV7670,
+    CAMERA_NT99141,
+    CAMERA_GC2145,
+    CAMERA_GC032A,
+    CAMERA_GC0308,
+    CAMERA_BF3005,
+    CAMERA_MODEL_MAX,
+    CAMERA_NONE,
+} camera_model_t;
+
+typedef enum {
+    OV2640_SCCB_ADDR   = 0x30,// 0x60 >> 1
+    OV5640_SCCB_ADDR   = 0x3C,// 0x78 >> 1
+    OV3660_SCCB_ADDR   = 0x3C,// 0x78 >> 1
+    OV7725_SCCB_ADDR   = 0x21,// 0x42 >> 1
+    OV7670_SCCB_ADDR   = 0x21,// 0x42 >> 1
+    NT99141_SCCB_ADDR  = 0x2A,// 0x54 >> 1
+    GC2145_SCCB_ADDR   = 0x3C,// 0x78 >> 1
+    GC032A_SCCB_ADDR   = 0x21,// 0x42 >> 1
+    GC0308_SCCB_ADDR   = 0x21,// 0x42 >> 1
+    BF3005_SCCB_ADDR   = 0x6E,
+} camera_sccb_addr_t;

 typedef enum {
    PIXFORMAT_RGB565,    // 2BPP/RGB565
@@ -58,6 +96,15 @@ typedef enum {
    FRAMESIZE_INVALID
 } framesize_t;

+typedef struct {
+    const camera_model_t model;
+    const char *name;
+    const camera_sccb_addr_t sccb_addr;
+    const camera_pid_t pid;
+    const framesize_t max_size;
+    const bool support_jpeg;
+} camera_sensor_info_t;
+
 typedef enum {
    ASPECT_RATIO_4X3,
    ASPECT_RATIO_3X2,
@@ -101,11 +148,13 @@ typedef struct {

 // Resolution table (in sensor.c)
 extern const resolution_info_t resolution[];
+// camera sensor table (in sensor.c)
+extern const camera_sensor_info_t camera_sensor[];

 typedef struct {
    uint8_t MIDH;
    uint8_t MIDL;
-    uint8_t PID;
+    uint16_t PID;
    uint8_t VER;
 } sensor_id_t;

@@ -190,4 +239,10 @@ typedef struct _sensor {
    int  (*set_xclk)            (sensor_t *sensor, int timer, int xclk);
 } sensor_t;

+camera_sensor_info_t *esp_camera_sensor_get_info(sensor_id_t *id);
+
+#ifdef __cplusplus
+}
+#endif
+
 #endif /* __SENSOR_H__ */
--- a/code/components/esp32-camera-master/driver/private_include/cam_hal.h
+++ b/code/components/esp32-camera-master/driver/private_include/cam_hal.h
@@ -0,0 +1,60 @@
+// Copyright 2010-2020 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+//
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#pragma once
+
+#include "esp_camera.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/**
+ * @brief Uninitialize the lcd_cam module
+ *
+ * @param handle Provide handle pointer to release resources
+ *
+ * @return
+ *     - ESP_OK Success
+ *     - ESP_FAIL Uninitialize fail
+ */
+esp_err_t cam_deinit(void);
+
+/**
+ * @brief Initialize the lcd_cam module
+ *
+ * @param config Configurations - see lcd_cam_config_t struct
+ *
+ * @return
+ *     - ESP_OK Success
+ *     - ESP_ERR_INVALID_ARG Parameter error
+ *     - ESP_ERR_NO_MEM No memory to initialize lcd_cam
+ *     - ESP_FAIL Initialize fail
+ */
+esp_err_t cam_init(const camera_config_t *config);
+
+esp_err_t cam_config(const camera_config_t *config, framesize_t frame_size, uint16_t sensor_pid);
+
+void cam_stop(void);
+
+void cam_start(void);
+
+camera_fb_t *cam_take(TickType_t timeout);
+
+void cam_give(camera_fb_t *dma_buffer);
+
+#ifdef __cplusplus
+}
+#endif
--- a/code/components/esp32-camera-master/driver/private_include/camera_common.h
+++ b/code/components/esp32-camera-master/driver/private_include/camera_common.h
@@ -1,49 +0,0 @@
-#pragma once
-
-#include <stdint.h>
-#include <stddef.h>
-#include <stdbool.h>
-#include "esp_err.h"
-#include "esp_intr_alloc.h"
-#include "freertos/FreeRTOS.h"
-#include "freertos/semphr.h"
-#include "freertos/task.h"
-#include "esp_camera.h"
-#include "sensor.h"
-
-#include "esp_system.h"
-#if ESP_IDF_VERSION_MAJOR >= 4 // IDF 4+
-#if CONFIG_IDF_TARGET_ESP32 // ESP32/PICO-D4
-#include "esp32/rom/lldesc.h"
-#else 
-#error Target CONFIG_IDF_TARGET is not supported
-#endif
-#else // ESP32 Before IDF 4.0
-#include "rom/lldesc.h"
-#endif
-
-typedef union {
-    struct {
-        uint8_t sample2;
-        uint8_t unused2;
-        uint8_t sample1;
-        uint8_t unused1;
-    };
-    uint32_t val;
-} dma_elem_t;
-
-typedef enum {
-    /* camera sends byte sequence: s1, s2, s3, s4, ...
-     * fifo receives: 00 s1 00 s2, 00 s2 00 s3, 00 s3 00 s4, ...
-     */
-    SM_0A0B_0B0C = 0,
-    /* camera sends byte sequence: s1, s2, s3, s4, ...
-     * fifo receives: 00 s1 00 s2, 00 s3 00 s4, ...
-     */
-    SM_0A0B_0C0D = 1,
-    /* camera sends byte sequence: s1, s2, s3, s4, ...
-     * fifo receives: 00 s1 00 00, 00 s2 00 00, 00 s3 00 00, ...
-     */
-    SM_0A00_0B00 = 3,
-} i2s_sampling_mode_t;
-
--- a/code/components/esp32-camera-master/driver/private_include/sccb.h
+++ b/code/components/esp32-camera-master/driver/private_include/sccb.h
@@ -10,6 +10,7 @@
 #define __SCCB_H__
 #include <stdint.h>
 int SCCB_Init(int pin_sda, int pin_scl);
+int SCCB_Deinit(void);
 uint8_t SCCB_Probe();
 uint8_t SCCB_Read(uint8_t slv_addr, uint8_t reg);
 uint8_t SCCB_Write(uint8_t slv_addr, uint8_t reg, uint8_t data);
--- a/code/components/esp32-camera-master/driver/private_include/xclk.h
+++ b/code/components/esp32-camera-master/driver/private_include/xclk.h
@@ -1,6 +1,8 @@
 #pragma once

-#include "camera_common.h"
+#include "esp_system.h"
+
+esp_err_t xclk_timer_conf(int ledc_timer, int xclk_freq_hz);

 esp_err_t camera_enable_out_clock();

--- a/code/components/esp32-camera-master/driver/sccb.c
+++ b/code/components/esp32-camera-master/driver/sccb.c
@@ -11,6 +11,7 @@
 #include <freertos/FreeRTOS.h>
 #include <freertos/task.h>
 #include "sccb.h"
+#include "sensor.h"
 #include <stdio.h>
 #include "sdkconfig.h"
 #if defined(ARDUINO_ARCH_ESP32) && defined(CONFIG_ARDUHAL_ESP_LOG)
@@ -24,24 +25,22 @@ static const char* TAG = "sccb";

 #include "driver/i2c.h"

-#define SCCB_FREQ               100000           /*!< I2C master frequency*/
-#define WRITE_BIT               I2C_MASTER_WRITE /*!< I2C master write */
-#define READ_BIT                I2C_MASTER_READ  /*!< I2C master read */
-#define ACK_CHECK_EN            0x1              /*!< I2C master will check ack from slave*/
-#define ACK_CHECK_DIS           0x0              /*!< I2C master will not check ack from slave */
-#define ACK_VAL                 0x0              /*!< I2C ack value */
-#define NACK_VAL                0x1              /*!< I2C nack value */
+#define SCCB_FREQ               CONFIG_SCCB_CLK_FREQ  /*!< I2C master frequency*/
+#define WRITE_BIT               I2C_MASTER_WRITE      /*!< I2C master write */
+#define READ_BIT                I2C_MASTER_READ       /*!< I2C master read */
+#define ACK_CHECK_EN            0x1                   /*!< I2C master will check ack from slave*/
+#define ACK_CHECK_DIS           0x0                   /*!< I2C master will not check ack from slave */
+#define ACK_VAL                 0x0                   /*!< I2C ack value */
+#define NACK_VAL                0x1                   /*!< I2C nack value */
 #if CONFIG_SCCB_HARDWARE_I2C_PORT1
 const int SCCB_I2C_PORT         = 1;
 #else
 const int SCCB_I2C_PORT         = 0;
 #endif
-static uint8_t ESP_SLAVE_ADDR   = 0x3c;

 int SCCB_Init(int pin_sda, int pin_scl)
 {
-    ESP_LOGI(TAG, "pin_sda %d pin_scl %d\n", pin_sda, pin_scl);
-    //log_i("SCCB_Init start");
+    ESP_LOGI(TAG, "pin_sda %d pin_scl %d", pin_sda, pin_scl);
    i2c_config_t conf;
    memset(&conf, 0, sizeof(i2c_config_t));
    conf.mode = I2C_MODE_MASTER;
@@ -56,10 +55,30 @@ int SCCB_Init(int pin_sda, int pin_scl)
    return 0;
 }

-uint8_t SCCB_Probe()
+int SCCB_Deinit(void)
+{
+    return i2c_driver_delete(SCCB_I2C_PORT);
+}
+
+uint8_t SCCB_Probe(void)
 {
    uint8_t slave_addr = 0x0;
-    while(slave_addr < 0x7f) {
+    // for (size_t i = 1; i < 0x80; i++) {
+    //     i2c_cmd_handle_t cmd = i2c_cmd_link_create();
+    //     i2c_master_start(cmd);
+    //     i2c_master_write_byte(cmd, ( i << 1 ) | WRITE_BIT, ACK_CHECK_EN);
+    //     i2c_master_stop(cmd);
+    //     esp_err_t ret = i2c_master_cmd_begin(SCCB_I2C_PORT, cmd, 1000 / portTICK_RATE_MS);
+    //     i2c_cmd_link_delete(cmd);
+    //     if( ret == ESP_OK) {
+    //         ESP_LOGW(TAG, "Found I2C Device at 0x%02X", i);
+    //     }
+    // }
+    for (size_t i = 0; i < CAMERA_MODEL_MAX; i++) {
+        if (slave_addr == camera_sensor[i].sccb_addr) {
+            continue;
+        }
+        slave_addr = camera_sensor[i].sccb_addr;
        i2c_cmd_handle_t cmd = i2c_cmd_link_create();
        i2c_master_start(cmd);
        i2c_master_write_byte(cmd, ( slave_addr << 1 ) | WRITE_BIT, ACK_CHECK_EN);
@@ -67,12 +86,10 @@ uint8_t SCCB_Probe()
        esp_err_t ret = i2c_master_cmd_begin(SCCB_I2C_PORT, cmd, 1000 / portTICK_RATE_MS);
        i2c_cmd_link_delete(cmd);
        if( ret == ESP_OK) {
-            ESP_SLAVE_ADDR = slave_addr;
-            return ESP_SLAVE_ADDR;
+            return slave_addr;
        }
-        slave_addr++;
    }
-    return ESP_SLAVE_ADDR;
+    return 0;
 }

 uint8_t SCCB_Read(uint8_t slv_addr, uint8_t reg)
--- a/code/components/esp32-camera-master/driver/sensor.c
+++ b/code/components/esp32-camera-master/driver/sensor.c
@@ -1,5 +1,20 @@
+#include <stdio.h>
 #include "sensor.h"

+const camera_sensor_info_t camera_sensor[CAMERA_MODEL_MAX] = {
+    // The sequence must be consistent with camera_model_t
+    {CAMERA_OV7725, "OV7725", OV7725_SCCB_ADDR, OV7725_PID, FRAMESIZE_VGA, false},
+    {CAMERA_OV2640, "OV2640", OV2640_SCCB_ADDR, OV2640_PID, FRAMESIZE_UXGA, true},
+    {CAMERA_OV3660, "OV3660", OV3660_SCCB_ADDR, OV3660_PID, FRAMESIZE_QXGA, true},
+    {CAMERA_OV5640, "OV5640", OV5640_SCCB_ADDR, OV5640_PID, FRAMESIZE_QSXGA, true},
+    {CAMERA_OV7670, "OV7670", OV7670_SCCB_ADDR, OV7670_PID, FRAMESIZE_VGA, false},
+    {CAMERA_NT99141, "NT99141", NT99141_SCCB_ADDR, NT99141_PID, FRAMESIZE_HD, true},
+    {CAMERA_GC2145, "GC2145", GC2145_SCCB_ADDR, GC2145_PID, FRAMESIZE_UXGA, false},
+    {CAMERA_GC032A, "GC032A", GC032A_SCCB_ADDR, GC032A_PID, FRAMESIZE_VGA, false},
+    {CAMERA_GC0308, "GC0308", GC0308_SCCB_ADDR, GC0308_PID, FRAMESIZE_VGA, false},
+    {CAMERA_BF3005, "BF3005", BF3005_SCCB_ADDR, BF3005_PID, FRAMESIZE_VGA, false},
+};
+
 const resolution_info_t resolution[FRAMESIZE_INVALID] = {
    {   96,   96, ASPECT_RATIO_1X1   }, /* 96x96 */
    {  160,  120, ASPECT_RATIO_4X3   }, /* QQVGA */
@@ -26,3 +41,13 @@ const resolution_info_t resolution[FRAMESIZE_INVALID] = {
    { 1088, 1920, ASPECT_RATIO_9X16  }, /* Portrait FHD   */
    { 2560, 1920, ASPECT_RATIO_4X3   }, /* QSXGA  */
 };
+
+camera_sensor_info_t *esp_camera_sensor_get_info(sensor_id_t *id)
+{
+    for (int i = 0; i < CAMERA_MODEL_MAX; i++) {
+        if (id->PID == camera_sensor[i].pid) {
+            return (camera_sensor_info_t *)&camera_sensor[i];
+        }
+    }
+    return NULL;
+}
--- a/code/components/esp32-camera-master/examples/CMakeLists.txt
+++ b/code/components/esp32-camera-master/examples/CMakeLists.txt
@@ -0,0 +1,9 @@
+# The following lines of boilerplate have to be in your project's
+# CMakeLists in this exact order for cmake to work correctly
+cmake_minimum_required(VERSION 3.5)
+
+set(EXTRA_COMPONENT_DIRS "../")
+
+add_compile_options(-fdiagnostics-color=always)
+include($ENV{IDF_PATH}/tools/cmake/project.cmake)
+project(camera_example)
--- a/code/components/esp32-camera-master/examples/main/CMakeLists.txt
+++ b/code/components/esp32-camera-master/examples/main/CMakeLists.txt
@@ -0,0 +1,3 @@
+set(COMPONENT_SRCS take_picture.c)
+set(COMPONENT_ADD_INCLUDEDIRS .)
+register_component()
--- a/code/components/esp32-camera-master/examples/main/component.mk
+++ b/code/components/esp32-camera-master/examples/main/component.mk
@@ -0,0 +1,5 @@
+#
+# "main" pseudo-component makefile.
+#
+# (Uses default behaviour of compiling all source files in directory, adding 'include' to include path.)
+
--- a/code/components/esp32-camera-master/examples/main/take_picture.c
+++ b/code/components/esp32-camera-master/examples/main/take_picture.c
@@ -29,7 +29,6 @@

 // ================================ CODE ======================================

-#include <esp_event.h>
 #include <esp_log.h>
 #include <esp_system.h>
 #include <nvs_flash.h>
@@ -41,6 +40,8 @@

 #include "esp_camera.h"

+#define BOARD_WROVER_KIT 1
+
 // WROVER-KIT PIN Map
 #ifdef BOARD_WROVER_KIT

@@ -113,11 +114,12 @@ static camera_config_t camera_config = {
    .ledc_timer = LEDC_TIMER_0,
    .ledc_channel = LEDC_CHANNEL_0,

-    .pixel_format = PIXFORMAT_JPEG, //YUV422,GRAYSCALE,RGB565,JPEG
-    .frame_size = FRAMESIZE_VGA,    //QQVGA-UXGA Do not use sizes above QVGA when not JPEG
+    .pixel_format = PIXFORMAT_RGB565, //YUV422,GRAYSCALE,RGB565,JPEG
+    .frame_size = FRAMESIZE_QVGA,    //QQVGA-UXGA Do not use sizes above QVGA when not JPEG

    .jpeg_quality = 12, //0-63 lower number means higher quality
-    .fb_count = 1       //if more than one, i2s runs in continuous mode. Use only with JPEG
+    .fb_count = 1,       //if more than one, i2s runs in continuous mode. Use only with JPEG
+    .grab_mode = CAMERA_GRAB_WHEN_EMPTY,
 };

 static esp_err_t init_camera()
@@ -135,7 +137,9 @@ static esp_err_t init_camera()

 void app_main()
 {
-    init_camera();
+    if(ESP_OK != init_camera()) {
+        return;
+    }

    while (1)
    {
@@ -144,7 +148,8 @@ void app_main()

        // use pic->buf to access the image
        ESP_LOGI(TAG, "Picture taken! Its size was: %zu bytes", pic->len);
+        esp_camera_fb_return(pic);

        vTaskDelay(5000 / portTICK_RATE_MS);
    }
-}
+}
--- a/code/components/esp32-camera-master/examples/sdkconfig.defaults
+++ b/code/components/esp32-camera-master/examples/sdkconfig.defaults
@@ -0,0 +1,17 @@
+CONFIG_ESP32_DEFAULT_CPU_FREQ_240=y
+CONFIG_ESP32S2_DEFAULT_CPU_FREQ_240=y
+CONFIG_ESP32S3_DEFAULT_CPU_FREQ_240=y
+
+CONFIG_ESPTOOLPY_FLASHSIZE_4MB=y
+CONFIG_PARTITION_TABLE_OFFSET=0x10000
+
+CONFIG_FREERTOS_HZ=1000
+CONFIG_ESPTOOLPY_FLASHFREQ_80M=y
+CONFIG_ESPTOOLPY_FLASHMODE_QIO=y
+
+CONFIG_SPIRAM_SUPPORT=y
+CONFIG_ESP32_SPIRAM_SUPPORT=y
+CONFIG_ESP32S2_SPIRAM_SUPPORT=y
+CONFIG_ESP32S3_SPIRAM_SUPPORT=y
+CONFIG_SPIRAM_SPEED_80M=y
+
--- a/code/components/esp32-camera-master/idf_component.yml
+++ b/code/components/esp32-camera-master/idf_component.yml
@@ -1,3 +1,5 @@
-version: "1.0.0"
-description: This package hosts ESP32 compatible driver for OV2640 image sensors. Additionally it provides a few tools, which allow converting the captured frame data to the more common BMP and JPEG formats.
-url: https://github.com/espressif/esp32-camera
+description: ESP32 compatible driver for OV2640, OV3660, OV5640, OV7670 and OV7725 image sensors.
+targets:
+  - esp32
+  - esp32s2
+  - esp32s3
--- a/code/components/esp32-camera-master/library.json
+++ b/code/components/esp32-camera-master/library.json
@@ -1,6 +1,6 @@
 {
  "name": "esp32-camera",
-  "version": "1.0.0",
+  "version": "2.0.0",
  "keywords": "esp32, camera, espressif, esp32-cam",
  "description": "ESP32 compatible driver for OV2640, OV3660, OV5640, OV7670 and OV7725 image sensors.",
  "repository": {
@@ -16,6 +16,7 @@
      "-Idriver/private_include",
      "-Iconversions/private_include",
      "-Isensors/private_include",
+      "-Itarget/private_include",
      "-fno-rtti"
    ],
    "includeDir": ".",
--- a/code/components/esp32-camera-master/sensors/bf3005.c
+++ b/code/components/esp32-camera-master/sensors/bf3005.c
@@ -0,0 +1,541 @@
+/*
+ * This file is part of the OpenMV project.
+ * Copyright (c) 2013/2014 Ibrahim Abdelkader <i.abdalkader@gmail.com>
+ * This work is licensed under the MIT license, see the file LICENSE for details.
+ *
+ * BF3005 driver.
+ * 
+ * Copyright 2015-2021 Espressif Systems (Shanghai) PTE LTD
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+
+ *     http://www.apache.org/licenses/LICENSE-2.0
+ * 
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ *
+ */
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+#include "sccb.h"
+#include "xclk.h"
+#include "bf3005.h"
+#include "bf3005_regs.h"
+#include "freertos/FreeRTOS.h"
+#include "freertos/task.h"
+
+#if defined(ARDUINO_ARCH_ESP32) && defined(CONFIG_ARDUHAL_ESP_LOG)
+#include "esp32-hal-log.h"
+#else
+#include "esp_log.h"
+static const char* TAG = "bf3005";
+#endif
+
+static const uint8_t default_regs[][2] = {
+  {0x12, 0x40}, //soft reset
+  {0xff, 0xff}, //delay 
+  {0xff, 0xff}, //delay 
+  {0xff, 0xff}, //delay 
+  {0xff, 0xff}, //delay 
+  {0x13, 0x10},	
+  {0x8c, 0x00},
+  {0x8d, 0x64},
+  {0x87, 0x10},
+  {0x13, 0x17},
+  {0x00, 0x20},
+  {0x01, 0x1a},
+  {0x02, 0x22},
+  {0x09, 0x03},
+  {0x0c, 0x80},
+  {0x0d, 0x24},
+  {0x0e, 0x21},
+  {0x0f, 0x28},
+  {0x11, 0x08},
+  {0x15, 0x10}, // 0X10
+  {0x16, 0x03},
+  {0x1e, 0x30},
+  {0x20, 0x8a},
+  {0x21, 0x03},
+  {0x23, 0x55},
+  {0x24, 0x68},
+  {0x25, 0x78},
+  {0x2a, 0x00},
+  {0x2b, 0x00},
+  {0x2d, 0x4f},
+  {0x2e, 0x98},
+  {0x2f, 0x04},
+  {0x30, 0xad},
+  {0x31, 0x17},
+  {0x32, 0x6e},
+  {0x33, 0x20},
+  {0x35, 0xa6},
+  {0x3b, 0x00},
+  {0x3e, 0x00},
+  {0x3f, 0xA8},
+  {0x40, 0x38},
+  {0x41, 0x32},
+  {0x42, 0x2b},
+  {0x43, 0x26},
+  {0x44, 0x1a},
+  {0x45, 0x16},
+  {0x46, 0x10},
+  {0x47, 0x0f},
+  {0x48, 0x0c},
+  {0x49, 0x0a},
+  {0x4b, 0x09},
+  {0x4c, 0x08},
+  {0x4d, 0x3c},
+  {0x4e, 0x06},
+  {0x4f, 0x05},
+  {0x50, 0x03},
+  {0x51, 0x25},
+  {0x52, 0x88},
+  {0x53, 0x03},
+  {0x63, 0x20},
+  {0x64, 0x02},
+  {0x65, 0xa6},
+  {0x66, 0xb6},
+  {0x69, 0x00},
+  {0x70, 0xFF},
+  {0x71, 0xa6},
+  {0x72, 0x2f},
+  {0x73, 0x2f},
+  {0x74, 0x2F},
+  {0x75, 0x0e},
+  {0x76, 0x1e},
+  {0x77, 0x00},
+  {0x78, 0x1e},
+  {0x79, 0x8a},
+  {0x7d, 0xe2},
+  {0x80, 0x44},
+  {0x81, 0x00},
+  {0x82, 0x18},
+  {0x83, 0x1b},
+  {0x84, 0x24},
+  {0x85, 0x2a},
+  {0x86, 0x4f},
+  {0x89, 0x82}, //0x82
+  {0x8b, 0x02},
+  {0x8e, 0x03},
+  {0x8f, 0xFC},
+  {0x9d, 0x4d},
+  {0x9e, 0x41},
+  {0xa1, 0x21},
+  {0xa2, 0x12},
+  {0xa3, 0x32},
+  {0xa4, 0x05},
+  {0xa5, 0x32},
+  {0xa6, 0x04},
+  {0xa7, 0x7f},
+  {0xa8, 0x7f},
+  {0xa9, 0x21},
+  {0xaa, 0x21},
+  {0xab, 0x21},
+  {0xac, 0x0a},
+  {0xad, 0xf0},
+  {0xae, 0xff},
+  {0xaf, 0x1d},
+  {0xb0, 0x94},
+  {0xb1, 0xc0},
+  {0xb2, 0xc0},
+  {0xd2, 0x30},
+  {0xe0, 0x0d},
+  {0xe1, 0x44},
+  {0xe7, 0x7c},
+  {0xe8, 0x89},
+  {0xe9, 0x01},
+  {0xea, 0x01},
+  {0xf0, 0x01},
+  {0xf3, 0x49},
+  {0xf4, 0xff},
+  {0xf5, 0x01},
+  {0xf6, 0xf2},
+  {0xf7, 0x6f},
+  {0x1b, 0x80},
+  {0x00, 0x00},
+};
+
+static int get_reg(sensor_t *sensor, int reg, int mask)
+{
+    int ret = SCCB_Read(sensor->slv_addr, reg & 0xFF);
+    if(ret > 0){
+        ret &= mask;
+    }
+    return ret;
+}
+
+static int set_reg(sensor_t *sensor, int reg, int mask, int value)
+{
+    int ret = 0;
+    ret = SCCB_Read(sensor->slv_addr, reg & 0xFF);
+    if(ret < 0){
+        return ret;
+    }
+    value = (ret & ~mask) | (value & mask);
+    ret = SCCB_Write(sensor->slv_addr, reg & 0xFF, value);
+    return ret;
+}
+
+static int set_reg_bits(sensor_t *sensor, uint8_t reg, uint8_t offset, uint8_t length, uint8_t value)
+{
+    int ret = 0;
+    ret = SCCB_Read(sensor->slv_addr, reg);
+    if(ret < 0){
+        return ret;
+    }
+    uint8_t mask = ((1 << length) - 1) << offset;
+    value = (ret & ~mask) | ((value << offset) & mask);
+    ret = SCCB_Write(sensor->slv_addr, reg & 0xFF, value);
+    return ret;
+}
+
+static int get_reg_bits(sensor_t *sensor, uint8_t reg, uint8_t offset, uint8_t length)
+{
+    int ret = 0;
+    ret = SCCB_Read(sensor->slv_addr, reg);
+    if(ret < 0){
+        return ret;
+    }
+    uint8_t mask = ((1 << length) - 1) << offset;
+    return (ret & mask) >> offset;
+}
+
+
+static int reset(sensor_t *sensor)
+{
+    int i=0;
+    const uint8_t (*regs)[2];
+
+    // Write default regsiters
+    for (i=0, regs = default_regs; regs[i][0]; i++) {
+        SCCB_Write(sensor->slv_addr, regs[i][0], regs[i][1]);
+    }
+
+    // Delay
+    vTaskDelay(50 / portTICK_PERIOD_MS);
+
+    return 0;
+}
+
+static int set_pixformat(sensor_t *sensor, pixformat_t pixformat)
+{
+    int ret=0;
+    sensor->pixformat = pixformat;
+
+    switch (pixformat) {
+    case PIXFORMAT_RGB565:
+        set_reg_bits(sensor, 0x12, 2, 1, 1);
+        break;
+    case PIXFORMAT_RAW:
+        set_reg_bits(sensor, 0x12, 0, 3, 0x4);
+        break;
+    case PIXFORMAT_YUV422:
+    case PIXFORMAT_GRAYSCALE:
+       set_reg_bits(sensor, 0x12, 2, 1, 0);
+        break;
+    default:
+        return -1;
+    }
+
+    // Delay
+    vTaskDelay(30 / portTICK_PERIOD_MS);
+
+    return ret;
+}
+
+static int set_framesize(sensor_t *sensor, framesize_t framesize)
+{
+    int ret=0;
+    if (framesize > FRAMESIZE_VGA) {
+        return -1;
+    }
+    uint16_t w = resolution[framesize].width;
+    uint16_t h = resolution[framesize].height;
+    // uint8_t reg = SCCB_Read(sensor->slv_addr, COM7);
+
+    sensor->status.framesize = framesize;
+
+    // Write MSBs
+    ret |= SCCB_Write(sensor->slv_addr, 0x17, 0);
+    ret |= SCCB_Write(sensor->slv_addr, 0x18, w>>2);
+
+    ret |= SCCB_Write(sensor->slv_addr, 0x19, 0);
+    ret |= SCCB_Write(sensor->slv_addr, 0x1a, h>>2);
+
+    // Write LSBs
+    ret |= SCCB_Write(sensor->slv_addr, 0x03, 0);
+    printf("%s %d\r\n", __func__, __LINE__);
+    if((w<=320)&&(h<=240))
+    {
+        printf("%s %d\r\n", __func__, __LINE__);
+        // Enable auto-scaling/zooming factors
+        //ret |= SCCB_Write(sensor->slv_addr, 0x12, 0x50);
+        set_reg_bits(sensor, 0x12, 4, 1, 1);
+
+        ret |= SCCB_Write(sensor->slv_addr, 0x17, (80-w/4));
+        ret |= SCCB_Write(sensor->slv_addr, 0x18, (80+w/4));
+
+        ret |= SCCB_Write(sensor->slv_addr, 0x19, (60-h/4));
+
+        ret |= SCCB_Write(sensor->slv_addr, 0x1a, (60+h/4));
+        ret |= SCCB_Write(sensor->slv_addr, 0x03, 0);
+
+    } else if((w<=640)&&(h<=480))
+    	{
+       // Enable auto-scaling/zooming factors
+        //ret |= SCCB_Write(sensor->slv_addr, 0x12, 0x40);
+        set_reg_bits(sensor, 0x12, 4, 1, 0);
+
+        ret |= SCCB_Write(sensor->slv_addr, 0x17, (80-w/8));
+        ret |= SCCB_Write(sensor->slv_addr, 0x18, (80+w/8));
+
+        ret |= SCCB_Write(sensor->slv_addr, 0x19, (60-h/8));
+
+        ret |= SCCB_Write(sensor->slv_addr, 0x1a, (60+h/8));
+        ret |= SCCB_Write(sensor->slv_addr, 0x03, 0);
+    }
+
+    // Delay
+    vTaskDelay(30 / portTICK_PERIOD_MS);
+
+    return ret;
+}
+
+static int set_colorbar(sensor_t *sensor, int value)
+{
+    int ret=0;
+    sensor->status.colorbar = value;
+
+    ret |= SCCB_Write(sensor->slv_addr, 0xb9, value);
+
+    return ret;
+}
+
+static int set_whitebal(sensor_t *sensor, int enable)
+{
+    if(set_reg_bits(sensor, 0x13, 1, 1, enable) >= 0){
+        sensor->status.awb = !!enable;
+    }
+    return sensor->status.awb;
+}
+
+
+static int set_gain_ctrl(sensor_t *sensor, int enable)
+{
+    if(set_reg_bits(sensor, 0x13, 2, 1, enable) >= 0){
+        sensor->status.agc = !!enable;
+    }
+    return sensor->status.agc;
+}
+
+
+static int set_exposure_ctrl(sensor_t *sensor, int enable)
+{
+    if(set_reg_bits(sensor, 0x13, 0, 1, enable) >= 0){
+        sensor->status.aec = !!enable;
+    }
+    return sensor->status.aec;
+}
+
+static int set_hmirror(sensor_t *sensor, int enable)
+{
+    if(set_reg_bits(sensor, 0x1e, 5, 1, enable) >= 0){
+        sensor->status.hmirror = !!enable;
+    }
+    return sensor->status.hmirror;
+}
+
+static int set_vflip(sensor_t *sensor, int enable)
+{
+    if(set_reg_bits(sensor, 0x1e, 4, 1, enable) >= 0){
+        sensor->status.vflip = !!enable;
+    }
+    return sensor->status.vflip;
+}
+
+static int set_raw_gma_dsp(sensor_t *sensor, int enable)
+{
+    int ret = 0;
+    ret = set_reg_bits(sensor, 0xf1, 1, 1, !enable);
+    if (ret == 0) {
+        ESP_LOGD(TAG, "Set raw_gma to: %d", !enable);
+        sensor->status.raw_gma = !enable;
+    }
+    return ret;
+}
+
+
+static int set_lenc_dsp(sensor_t *sensor, int enable)
+{
+    int ret = 0;
+    ret = set_reg_bits(sensor, 0xf1, 0, 1, !enable);
+    if (ret == 0) {
+        ESP_LOGD(TAG, "Set lenc to: %d", !enable);
+        sensor->status.lenc = !enable;
+    }
+    return ret;
+}
+
+static int set_agc_gain(sensor_t *sensor, int option)
+{
+    int ret = 0;
+    ret = set_reg_bits(sensor, 0x13, 4, 1, !!option);
+    if (ret == 0) {
+        ESP_LOGD(TAG, "Set gain to: %d", !!option);
+        sensor->status.agc_gain = !!option;
+    }
+    return ret;
+}
+
+static int set_awb_gain_dsp(sensor_t *sensor, int value)
+{
+    int ret = 0;
+    ret = SCCB_Write(sensor->slv_addr, 0xa6, value);
+    if (ret == 0) {
+        ESP_LOGD(TAG, "Set awb gain threthold to: %d", value);
+        sensor->status.awb_gain = value;
+    }
+    return ret;
+}
+
+static int set_brightness(sensor_t *sensor, int level)
+{
+    int ret = 0;
+    ret = SCCB_Write(sensor->slv_addr, 0x55, level);
+    if (ret == 0) {
+        ESP_LOGD(TAG, "Set brightness to: %d", level);
+        sensor->status.brightness = level;
+    }
+    return ret;
+}
+
+static int set_contrast(sensor_t *sensor, int level)
+{
+    int ret = 0;
+    ret = SCCB_Write(sensor->slv_addr, 0x56, level);
+    if (ret == 0) {
+        ESP_LOGD(TAG, "Set contrast to: %d", level);
+        sensor->status.contrast = level;
+    }
+    return ret;
+}
+
+static int set_sharpness(sensor_t *sensor, int level)
+{
+    int ret = 0;
+    ret = SCCB_Write(sensor->slv_addr, 0x70, level);
+    if (ret == 0) {
+        ESP_LOGD(TAG, "Set sharpness to: %d", level);
+        sensor->status.sharpness = level;
+    }
+    return ret;
+}
+
+static int init_status(sensor_t *sensor)
+{
+    sensor->status.brightness = SCCB_Read(sensor->slv_addr, 0x55);
+    sensor->status.contrast = SCCB_Read(sensor->slv_addr, 0x56);
+    sensor->status.saturation = 0;
+    sensor->status.ae_level = 0;
+    
+    sensor->status.gainceiling = SCCB_Read(sensor->slv_addr, 0x87);
+    sensor->status.awb = get_reg_bits(sensor, 0x13, 1, 1);
+    sensor->status.awb_gain = SCCB_Read(sensor->slv_addr, 0xa6);
+    sensor->status.aec = get_reg_bits(sensor, 0x13, 0, 1);
+
+    sensor->status.agc = get_reg_bits(sensor, 0x13, 2, 1);
+    
+    sensor->status.raw_gma = get_reg_bits(sensor, 0xf1, 1, 1);
+    sensor->status.lenc = get_reg_bits(sensor, 0xf1, 0, 1);
+    sensor->status.hmirror = get_reg_bits(sensor, 0x1e, 5, 1);
+    sensor->status.vflip = get_reg_bits(sensor, 0x1e, 4, 1);
+    
+    sensor->status.colorbar = SCCB_Read(sensor->slv_addr, 0xb9);
+    sensor->status.sharpness = SCCB_Read(sensor->slv_addr, 0x70);
+    
+    return 0;
+}
+
+static int set_dummy(sensor_t *sensor, int val){ return -1; }
+static int set_gainceiling_dummy(sensor_t *sensor, gainceiling_t val){ return -1; }
+static int set_res_raw(sensor_t *sensor, int startX, int startY, int endX, int endY, int offsetX, int offsetY, int totalX, int totalY, int outputX, int outputY, bool scale, bool binning){return -1;}
+static int _set_pll(sensor_t *sensor, int bypass, int multiplier, int sys_div, int root_2x, int pre_div, int seld5, int pclk_manual, int pclk_div){return -1;}
+
+static int set_xclk(sensor_t *sensor, int timer, int xclk)
+{
+    int ret = 0;
+    sensor->xclk_freq_hz = xclk * 1000000U;
+    ret = xclk_timer_conf(timer, sensor->xclk_freq_hz);
+    return ret;
+}
+
+int bf3005_detect(int slv_addr, sensor_id_t *id)
+{
+    if (BF3005_SCCB_ADDR == slv_addr) {
+        uint16_t PID = SCCB_Read(slv_addr, 0xFC);
+        if (BF3005_PID == PID) {
+            id->PID = PID;
+            id->VER = SCCB_Read(slv_addr, 0xFD);
+            id->MIDL = SCCB_Read(slv_addr, 0xFC);
+            id->MIDH = SCCB_Read(slv_addr, 0xFD);
+            return PID;
+        } else {
+            ESP_LOGI(TAG, "Mismatch PID=0x%x", PID);
+        }
+    }
+    return 0;
+}
+
+int bf3005_init(sensor_t *sensor)
+{
+    // Set function pointers
+    sensor->reset = reset;
+    sensor->init_status = init_status;
+    sensor->set_pixformat = set_pixformat;
+    sensor->set_framesize = set_framesize;
+    sensor->set_brightness = set_brightness;
+    sensor->set_contrast = set_contrast;
+
+    sensor->set_colorbar = set_colorbar;
+
+    sensor->set_gain_ctrl = set_gain_ctrl;
+    sensor->set_exposure_ctrl = set_exposure_ctrl;
+    sensor->set_hmirror = set_hmirror;
+    sensor->set_vflip = set_vflip;
+
+    sensor->set_whitebal = set_whitebal;
+
+    sensor->set_awb_gain = set_awb_gain_dsp;
+    sensor->set_agc_gain = set_agc_gain;
+    
+    sensor->set_raw_gma = set_raw_gma_dsp;
+    sensor->set_lenc = set_lenc_dsp;
+
+    sensor->set_sharpness = set_sharpness;
+    //not supported
+    sensor->set_saturation= set_dummy;
+    sensor->set_denoise = set_dummy;
+    sensor->set_quality = set_dummy;
+    sensor->set_special_effect = set_dummy;
+    sensor->set_wb_mode = set_dummy;
+    sensor->set_ae_level = set_dummy;
+    sensor->set_gainceiling = set_gainceiling_dummy;
+
+
+    sensor->get_reg = get_reg;
+    sensor->set_reg = set_reg;
+    sensor->set_res_raw = set_res_raw;
+    sensor->set_pll = _set_pll;
+    sensor->set_xclk = set_xclk;
+    
+    ESP_LOGD(TAG, "BF3005 Attached");
+
+    return 0;
+}
--- a/code/components/esp32-camera-master/sensors/gc0308.c
+++ b/code/components/esp32-camera-master/sensors/gc0308.c
@@ -0,0 +1,467 @@
+// Copyright 2015-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+#include "freertos/FreeRTOS.h"
+#include "freertos/task.h"
+#include "sccb.h"
+#include "gc0308.h"
+#include "gc0308_regs.h"
+#include "gc0308_settings.h"
+
+#if defined(ARDUINO_ARCH_ESP32) && defined(CONFIG_ARDUHAL_ESP_LOG)
+#include "esp32-hal-log.h"
+#else
+#include "esp_log.h"
+static const char *TAG = "gc0308";
+#endif
+
+#define H8(v) ((v)>>8)
+#define L8(v) ((v)&0xff)
+
+//#define REG_DEBUG_ON
+
+static int read_reg(uint8_t slv_addr, const uint16_t reg)
+{
+    int ret = SCCB_Read(slv_addr, reg);
+#ifdef REG_DEBUG_ON
+    if (ret < 0) {
+        ESP_LOGE(TAG, "READ REG 0x%04x FAILED: %d", reg, ret);
+    }
+#endif
+    return ret;
+}
+
+static int write_reg(uint8_t slv_addr, const uint16_t reg, uint8_t value)
+{
+    int ret = 0;
+#ifndef REG_DEBUG_ON
+    ret = SCCB_Write(slv_addr, reg, value);
+#else
+    int old_value = read_reg(slv_addr, reg);
+    if (old_value < 0) {
+        return old_value;
+    }
+    if ((uint8_t)old_value != value) {
+        ESP_LOGI(TAG, "NEW REG 0x%04x: 0x%02x to 0x%02x", reg, (uint8_t)old_value, value);
+        ret = SCCB_Write(slv_addr, reg, value);
+    } else {
+        ESP_LOGD(TAG, "OLD REG 0x%04x: 0x%02x", reg, (uint8_t)old_value);
+        ret = SCCB_Write(slv_addr, reg, value);//maybe not?
+    }
+    if (ret < 0) {
+        ESP_LOGE(TAG, "WRITE REG 0x%04x FAILED: %d", reg, ret);
+    }
+#endif
+    return ret;
+}
+
+static int check_reg_mask(uint8_t slv_addr, uint16_t reg, uint8_t mask)
+{
+    return (read_reg(slv_addr, reg) & mask) == mask;
+}
+
+static int set_reg_bits(uint8_t slv_addr, uint16_t reg, uint8_t offset, uint8_t mask, uint8_t value)
+{
+    int ret = 0;
+    uint8_t c_value, new_value;
+    ret = read_reg(slv_addr, reg);
+    if (ret < 0) {
+        return ret;
+    }
+    c_value = ret;
+    new_value = (c_value & ~(mask << offset)) | ((value & mask) << offset);
+    ret = write_reg(slv_addr, reg, new_value);
+    return ret;
+}
+
+static int write_regs(uint8_t slv_addr, const uint16_t (*regs)[2])
+{
+    int i = 0, ret = 0;
+    while (!ret && regs[i][0] != REGLIST_TAIL) {
+        if (regs[i][0] == REG_DLY) {
+            vTaskDelay(regs[i][1] / portTICK_PERIOD_MS);
+        } else {
+            ret = write_reg(slv_addr, regs[i][0], regs[i][1]);
+        }
+        i++;
+    }
+    return ret;
+}
+
+static void print_regs(uint8_t slv_addr)
+{
+#ifdef DEBUG_PRINT_REG
+    ESP_LOGI(TAG, "REG list look ======================");
+    for (size_t i = 0xf0; i <= 0xfe; i++) {
+        ESP_LOGI(TAG, "reg[0x%02x] = 0x%02x", i, read_reg(slv_addr, i));
+    }
+    ESP_LOGI(TAG, "\npage 0 ===");
+    write_reg(slv_addr, 0xfe, 0x00); // page 0
+    for (size_t i = 0x03; i <= 0xa2; i++) {
+        ESP_LOGI(TAG, "p0 reg[0x%02x] = 0x%02x", i, read_reg(slv_addr, i));
+    }
+
+    ESP_LOGI(TAG, "\npage 3 ===");
+    write_reg(slv_addr, 0xfe, 0x03); // page 3
+    for (size_t i = 0x01; i <= 0x43; i++) {
+        ESP_LOGI(TAG, "p3 reg[0x%02x] = 0x%02x", i, read_reg(slv_addr, i));
+    }
+#endif
+}
+
+static int reset(sensor_t *sensor)
+{
+    int ret = 0;
+    // Software Reset: clear all registers and reset them to their default values
+    ret = write_reg(sensor->slv_addr, RESET_RELATED, 0xf0);
+    if (ret) {
+        ESP_LOGE(TAG, "Software Reset FAILED!");
+        return ret;
+    }
+    vTaskDelay(100 / portTICK_PERIOD_MS);
+    ret = write_regs(sensor->slv_addr, gc0308_sensor_default_regs);
+    if (ret == 0) {
+        ESP_LOGD(TAG, "Camera defaults loaded");
+        vTaskDelay(100 / portTICK_PERIOD_MS);
+        write_reg(sensor->slv_addr, 0xfe, 0x00);
+#ifdef CONFIG_IDF_TARGET_ESP32
+        set_reg_bits(sensor->slv_addr, 0x28, 4, 0x07, 1);  //frequency division for esp32, ensure pclk <= 15MHz
+#endif
+    }
+    return ret;
+}
+
+static int set_pixformat(sensor_t *sensor, pixformat_t pixformat)
+{
+    int ret = 0;
+
+    switch (pixformat) {
+    case PIXFORMAT_RGB565:
+        write_reg(sensor->slv_addr, 0xfe, 0x00);
+        ret = set_reg_bits(sensor->slv_addr, 0x24, 0, 0x0f, 6);  //RGB565
+        break;
+
+    case PIXFORMAT_YUV422:
+        write_reg(sensor->slv_addr, 0xfe, 0x00);
+        ret = set_reg_bits(sensor->slv_addr, 0x24, 0, 0x0f, 2); //yuv422 Y Cb Y Cr
+        break;
+    default:
+        ESP_LOGW(TAG, "unsupport format");
+        ret = -1;
+        break;
+    }
+
+    if (ret == 0) {
+        sensor->pixformat = pixformat;
+        ESP_LOGD(TAG, "Set pixformat to: %u", pixformat);
+    }
+    return ret;
+}
+
+static int set_framesize(sensor_t *sensor, framesize_t framesize)
+{
+    int ret = 0;
+    if (framesize > FRAMESIZE_VGA) {
+        ESP_LOGW(TAG, "Invalid framesize: %u", framesize);
+        framesize = FRAMESIZE_VGA;
+    }
+    sensor->status.framesize = framesize;
+    uint16_t w = resolution[framesize].width;
+    uint16_t h = resolution[framesize].height;
+    uint16_t row_s = (resolution[FRAMESIZE_VGA].height - h) / 2;
+    uint16_t col_s = (resolution[FRAMESIZE_VGA].width - w) / 2;
+    (void)row_s;
+    (void)col_s;
+
+#if CONFIG_GC_SENSOR_SUBSAMPLE_MODE
+    struct subsample_cfg {
+        uint16_t ratio_numerator;
+        uint16_t ratio_denominator;
+        uint8_t reg0x54;
+        uint8_t reg0x56;
+        uint8_t reg0x57;
+        uint8_t reg0x58;
+        uint8_t reg0x59;
+    };
+    const struct subsample_cfg subsample_cfgs[] = { // define some subsample ratio
+        {84, 420, 0x55, 0x00, 0x00, 0x00, 0x00}, //1/5
+        {105, 420, 0x44, 0x00, 0x00, 0x00, 0x00},//1/4
+        {140, 420, 0x33, 0x00, 0x00, 0x00, 0x00},//1/3
+        {210, 420, 0x22, 0x00, 0x00, 0x00, 0x00},//1/2
+        {240, 420, 0x77, 0x02, 0x46, 0x02, 0x46},//4/7
+        {252, 420, 0x55, 0x02, 0x04, 0x02, 0x04},//3/5
+        {280, 420, 0x33, 0x02, 0x00, 0x02, 0x00},//2/3
+        {420, 420, 0x11, 0x00, 0x00, 0x00, 0x00},//1/1
+    };
+    uint16_t win_w = 640;
+    uint16_t win_h = 480;
+    const struct subsample_cfg *cfg = NULL;
+    /**
+     * Strategy: try to keep the maximum perspective
+     */
+    for (size_t i = 0; i < sizeof(subsample_cfgs) / sizeof(struct subsample_cfg); i++) {
+        cfg = &subsample_cfgs[i];
+        if ((win_w * cfg->ratio_numerator / cfg->ratio_denominator >= w) && (win_h * cfg->ratio_numerator / cfg->ratio_denominator >= h)) {
+            win_w = w * cfg->ratio_denominator / cfg->ratio_numerator;
+            win_h = h * cfg->ratio_denominator / cfg->ratio_numerator;
+            row_s = (resolution[FRAMESIZE_VGA].height - win_h) / 2;
+            col_s = (resolution[FRAMESIZE_VGA].width - win_w) / 2;
+            ESP_LOGI(TAG, "subsample win:%dx%d, ratio:%f", win_w, win_h, (float)cfg->ratio_numerator / (float)cfg->ratio_denominator);
+            break;
+        }
+    }
+
+    write_reg(sensor->slv_addr, 0xfe, 0x00);
+
+    write_reg(sensor->slv_addr, 0x05, H8(row_s));
+    write_reg(sensor->slv_addr, 0x06, L8(row_s));
+    write_reg(sensor->slv_addr, 0x07, H8(col_s));
+    write_reg(sensor->slv_addr, 0x08, L8(col_s));
+    write_reg(sensor->slv_addr, 0x09, H8(win_h + 8));
+    write_reg(sensor->slv_addr, 0x0a, L8(win_h + 8));
+    write_reg(sensor->slv_addr, 0x0b, H8(win_w + 8));
+    write_reg(sensor->slv_addr, 0x0c, L8(win_w + 8));
+
+    write_reg(sensor->slv_addr, 0xfe, 0x01);
+    set_reg_bits(sensor->slv_addr, 0x53, 7, 0x01, 1);
+    set_reg_bits(sensor->slv_addr, 0x55, 0, 0x01, 1);
+    write_reg(sensor->slv_addr, 0x54, cfg->reg0x54);
+    write_reg(sensor->slv_addr, 0x56, cfg->reg0x56);
+    write_reg(sensor->slv_addr, 0x57, cfg->reg0x57);
+    write_reg(sensor->slv_addr, 0x58, cfg->reg0x58);
+    write_reg(sensor->slv_addr, 0x59, cfg->reg0x59);
+
+    write_reg(sensor->slv_addr, 0xfe, 0x00);
+
+#elif CONFIG_GC_SENSOR_WINDOWING_MODE
+    write_reg(sensor->slv_addr, 0xfe, 0x00);
+
+    write_reg(sensor->slv_addr, 0xf7, col_s / 4);
+    write_reg(sensor->slv_addr, 0xf8, row_s / 4);
+    write_reg(sensor->slv_addr, 0xf9, (col_s + h) / 4);
+    write_reg(sensor->slv_addr, 0xfa, (row_s + w) / 4);
+
+    write_reg(sensor->slv_addr, 0x05, H8(row_s));
+    write_reg(sensor->slv_addr, 0x06, L8(row_s));
+    write_reg(sensor->slv_addr, 0x07, H8(col_s));
+    write_reg(sensor->slv_addr, 0x08, L8(col_s));
+
+    write_reg(sensor->slv_addr, 0x09, H8(h + 8));
+    write_reg(sensor->slv_addr, 0x0a, L8(h + 8));
+    write_reg(sensor->slv_addr, 0x0b, H8(w + 8));
+    write_reg(sensor->slv_addr, 0x0c, L8(w + 8));
+
+#endif
+    if (ret == 0) {
+        ESP_LOGD(TAG, "Set framesize to: %ux%u", w, h);
+    }
+    return 0;
+}
+
+static int set_contrast(sensor_t *sensor, int contrast)
+{
+    if (contrast != 0) {
+        write_reg(sensor->slv_addr, 0xfe, 0x00);
+        write_reg(sensor->slv_addr, 0xb3, contrast);
+    }
+    return 0;
+}
+
+static int set_global_gain(sensor_t *sensor, int gain_level)
+{
+    if (gain_level != 0) {
+        write_reg(sensor->slv_addr, 0xfe, 0x00);
+        write_reg(sensor->slv_addr, 0x50, gain_level);
+    }
+    return 0;
+}
+
+static int set_hmirror(sensor_t *sensor, int enable)
+{
+    int ret = 0;
+    sensor->status.hmirror = enable;
+    ret = write_reg(sensor->slv_addr, 0xfe, 0x00);
+    ret |= set_reg_bits(sensor->slv_addr, 0x14, 0, 0x01, enable != 0);
+    if (ret == 0) {
+        ESP_LOGD(TAG, "Set h-mirror to: %d", enable);
+    }
+    return ret;
+}
+
+static int set_vflip(sensor_t *sensor, int enable)
+{
+    int ret = 0;
+    sensor->status.vflip = enable;
+    ret = write_reg(sensor->slv_addr, 0xfe, 0x00);
+    ret |= set_reg_bits(sensor->slv_addr, 0x14, 1, 0x01, enable != 0);
+    if (ret == 0) {
+        ESP_LOGD(TAG, "Set v-flip to: %d", enable);
+    }
+    return ret;
+}
+
+static int set_colorbar(sensor_t *sensor, int enable)
+{
+    int ret = 0;
+    ret = write_reg(sensor->slv_addr, 0xfe, 0x00);
+    ret |= set_reg_bits(sensor->slv_addr, 0x2e, 0, 0x01, enable);
+    if (ret == 0) {
+        sensor->status.colorbar = enable;
+        ESP_LOGD(TAG, "Set colorbar to: %d", enable);
+    }
+    return ret;
+}
+
+static int get_reg(sensor_t *sensor, int reg, int mask)
+{
+    int ret = 0;
+    if (mask > 0xFF) {
+        ESP_LOGE(TAG, "mask should not more than 0xff");
+    } else {
+        ret = read_reg(sensor->slv_addr, reg);
+    }
+    if (ret > 0) {
+        ret &= mask;
+    }
+    return ret;
+}
+
+static int set_reg(sensor_t *sensor, int reg, int mask, int value)
+{
+    int ret = 0;
+    if (mask > 0xFF) {
+        ESP_LOGE(TAG, "mask should not more than 0xff");
+    } else {
+        ret = read_reg(sensor->slv_addr, reg);
+    }
+    if (ret < 0) {
+        return ret;
+    }
+    value = (ret & ~mask) | (value & mask);
+
+    if (mask > 0xFF) {
+
+    } else {
+        ret = write_reg(sensor->slv_addr, reg, value);
+    }
+    return ret;
+}
+
+static int init_status(sensor_t *sensor)
+{
+    write_reg(sensor->slv_addr, 0xfe, 0x00);
+    sensor->status.brightness = 0;
+    sensor->status.contrast = 0;
+    sensor->status.saturation = 0;
+    sensor->status.sharpness = 0;
+    sensor->status.denoise = 0;
+    sensor->status.ae_level = 0;
+    sensor->status.gainceiling = 0;
+    sensor->status.awb = 0;
+    sensor->status.dcw = 0;
+    sensor->status.agc = 0;
+    sensor->status.aec = 0;
+    sensor->status.hmirror = check_reg_mask(sensor->slv_addr, 0x14, 0x01);
+    sensor->status.vflip = check_reg_mask(sensor->slv_addr, 0x14, 0x02);
+    sensor->status.colorbar = 0;
+    sensor->status.bpc = 0;
+    sensor->status.wpc = 0;
+    sensor->status.raw_gma = 0;
+    sensor->status.lenc = 0;
+    sensor->status.quality = 0;
+    sensor->status.special_effect = 0;
+    sensor->status.wb_mode = 0;
+    sensor->status.awb_gain = 0;
+    sensor->status.agc_gain = 0;
+    sensor->status.aec_value = 0;
+    sensor->status.aec2 = 0;
+
+    print_regs(sensor->slv_addr);
+    return 0;
+}
+
+static int set_dummy(sensor_t *sensor, int val)
+{
+    ESP_LOGW(TAG, "Unsupported");
+    return -1;
+}
+static int set_gainceiling_dummy(sensor_t *sensor, gainceiling_t val)
+{
+    ESP_LOGW(TAG, "Unsupported");
+    return -1;
+}
+
+int gc0308_detect(int slv_addr, sensor_id_t *id)
+{
+    if (GC0308_SCCB_ADDR == slv_addr) {
+        write_reg(slv_addr, 0xfe, 0x00);
+        uint8_t PID = SCCB_Read(slv_addr, 0x00);
+        if (GC0308_PID == PID) {
+            id->PID = PID;
+            return PID;
+        } else {
+            ESP_LOGI(TAG, "Mismatch PID=0x%x", PID);
+        }
+    }
+    return 0;
+}
+
+int gc0308_init(sensor_t *sensor)
+{
+    sensor->init_status = init_status;
+    sensor->reset = reset;
+    sensor->set_pixformat = set_pixformat;
+    sensor->set_framesize = set_framesize;
+    sensor->set_contrast = set_contrast;
+    sensor->set_brightness = set_dummy;
+    sensor->set_saturation = set_dummy;
+    sensor->set_sharpness = set_dummy;
+    sensor->set_denoise = set_dummy;
+    sensor->set_gainceiling = set_gainceiling_dummy;
+    sensor->set_quality = set_dummy;
+    sensor->set_colorbar = set_colorbar;
+    sensor->set_whitebal = set_dummy;
+    sensor->set_gain_ctrl = set_global_gain;
+    sensor->set_exposure_ctrl = set_dummy;
+    sensor->set_hmirror = set_hmirror;
+    sensor->set_vflip = set_vflip;
+
+    sensor->set_aec2 = set_dummy;
+    sensor->set_awb_gain = set_dummy;
+    sensor->set_agc_gain = set_dummy;
+    sensor->set_aec_value = set_dummy;
+
+    sensor->set_special_effect = set_dummy;
+    sensor->set_wb_mode = set_dummy;
+    sensor->set_ae_level = set_dummy;
+
+    sensor->set_dcw = set_dummy;
+    sensor->set_bpc = set_dummy;
+    sensor->set_wpc = set_dummy;
+
+    sensor->set_raw_gma = set_dummy;
+    sensor->set_lenc = set_dummy;
+
+    sensor->get_reg = get_reg;
+    sensor->set_reg = set_reg;
+    sensor->set_res_raw = NULL;
+    sensor->set_pll = NULL;
+    sensor->set_xclk = NULL;
+
+    ESP_LOGD(TAG, "GC0308 Attached");
+    return 0;
+}
--- a/code/components/esp32-camera-master/sensors/gc032a.c
+++ b/code/components/esp32-camera-master/sensors/gc032a.c
@@ -0,0 +1,391 @@
+// Copyright 2015-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+#include <stdio.h>
+#include "freertos/FreeRTOS.h"
+#include "freertos/task.h"
+#include "sccb.h"
+#include "gc032a.h"
+#include "gc032a_regs.h"
+#include "gc032a_settings.h"
+
+#if defined(ARDUINO_ARCH_ESP32) && defined(CONFIG_ARDUHAL_ESP_LOG)
+#include "esp32-hal-log.h"
+#else
+#include "esp_log.h"
+static const char *TAG = "gc032a";
+#endif
+
+#define H8(v) ((v)>>8)
+#define L8(v) ((v)&0xff)
+
+//#define REG_DEBUG_ON
+
+static int read_reg(uint8_t slv_addr, const uint16_t reg)
+{
+    int ret = SCCB_Read(slv_addr, reg);
+#ifdef REG_DEBUG_ON
+    if (ret < 0) {
+        ESP_LOGE(TAG, "READ REG 0x%04x FAILED: %d", reg, ret);
+    }
+#endif
+    return ret;
+}
+
+static int write_reg(uint8_t slv_addr, const uint16_t reg, uint8_t value)
+{
+    int ret = 0;
+#ifndef REG_DEBUG_ON
+    ret = SCCB_Write(slv_addr, reg, value);
+#else
+    int old_value = read_reg(slv_addr, reg);
+    if (old_value < 0) {
+        return old_value;
+    }
+    if ((uint8_t)old_value != value) {
+        ESP_LOGI(TAG, "NEW REG 0x%04x: 0x%02x to 0x%02x", reg, (uint8_t)old_value, value);
+        ret = SCCB_Write(slv_addr, reg, value);
+    } else {
+        ESP_LOGD(TAG, "OLD REG 0x%04x: 0x%02x", reg, (uint8_t)old_value);
+        ret = SCCB_Write(slv_addr, reg, value);//maybe not?
+    }
+    if (ret < 0) {
+        ESP_LOGE(TAG, "WRITE REG 0x%04x FAILED: %d", reg, ret);
+    }
+#endif
+    return ret;
+}
+
+static int check_reg_mask(uint8_t slv_addr, uint16_t reg, uint8_t mask)
+{
+    return (read_reg(slv_addr, reg) & mask) == mask;
+}
+
+static void print_regs(uint8_t slv_addr)
+{
+#ifdef DEBUG_PRINT_REG
+    vTaskDelay(pdMS_TO_TICKS(100));
+    ESP_LOGI(TAG, "REG list look ======================");
+    for (size_t i = 0xf0; i <= 0xfe; i++) {
+        ESP_LOGI(TAG, "reg[0x%02x] = 0x%02x", i, read_reg(slv_addr, i));
+    }
+    ESP_LOGI(TAG, "\npage 0 ===");
+    write_reg(slv_addr, 0xfe, 0x00); // page 0
+    for (size_t i = 0x03; i <= 0x24; i++) {
+        ESP_LOGI(TAG, "p0 reg[0x%02x] = 0x%02x", i, read_reg(slv_addr, i));
+    }
+    for (size_t i = 0x40; i <= 0x95; i++) {
+        ESP_LOGI(TAG, "p0 reg[0x%02x] = 0x%02x", i, read_reg(slv_addr, i));
+    }
+    ESP_LOGI(TAG, "\npage 3 ===");
+    write_reg(slv_addr, 0xfe, 0x03); // page 3
+    for (size_t i = 0x01; i <= 0x43; i++) {
+        ESP_LOGI(TAG, "p3 reg[0x%02x] = 0x%02x", i, read_reg(slv_addr, i));
+    }
+#endif
+}
+
+static int set_reg_bits(uint8_t slv_addr, uint16_t reg, uint8_t offset, uint8_t mask, uint8_t value)
+{
+    int ret = 0;
+    uint8_t c_value, new_value;
+    ret = read_reg(slv_addr, reg);
+    if (ret < 0) {
+        return ret;
+    }
+    c_value = ret;
+    new_value = (c_value & ~(mask << offset)) | ((value & mask) << offset);
+    ret = write_reg(slv_addr, reg, new_value);
+    return ret;
+}
+
+static int write_regs(uint8_t slv_addr, const uint16_t (*regs)[2])
+{
+    int i = 0, ret = 0;
+    while (!ret && regs[i][0] != REGLIST_TAIL) {
+        if (regs[i][0] == REG_DLY) {
+            vTaskDelay(regs[i][1] / portTICK_PERIOD_MS);
+        } else {
+            ret = write_reg(slv_addr, regs[i][0], regs[i][1]);
+        }
+        i++;
+    }
+    return ret;
+}
+
+static int reset(sensor_t *sensor)
+{
+    int ret;
+    // Software Reset: clear all registers and reset them to their default values
+    ret = write_reg(sensor->slv_addr, RESET_RELATED, 0xf0);
+    if (ret) {
+        ESP_LOGE(TAG, "Software Reset FAILED!");
+        return ret;
+    }
+    vTaskDelay(100 / portTICK_PERIOD_MS);
+
+    ret = write_regs(sensor->slv_addr, gc032a_default_regs);
+    if (ret == 0) {
+        ESP_LOGD(TAG, "Camera defaults loaded");
+        vTaskDelay(100 / portTICK_PERIOD_MS);
+        write_reg(sensor->slv_addr, 0xfe, 0x00);
+        set_reg_bits(sensor->slv_addr, 0xf7, 1, 0x01, 1); // PLL_mode1:div2en
+        set_reg_bits(sensor->slv_addr, 0xf7, 7, 0x01, 1); // PLL_mode1:dvp mode
+        set_reg_bits(sensor->slv_addr, 0xf8, 0, 0x3f, 8); //PLL_mode2 :divx4
+        set_reg_bits(sensor->slv_addr, 0xfa, 4, 0x0f, 2); //vlk div mode :divide_by
+    }
+
+    return ret;
+}
+
+static int set_pixformat(sensor_t *sensor, pixformat_t pixformat)
+{
+    int ret = 0;
+    switch (pixformat) {
+    case PIXFORMAT_RGB565:
+        write_reg(sensor->slv_addr, 0xfe, 0x00);
+        ret = set_reg_bits(sensor->slv_addr, 0x44, 0, 0x1f, 6);  //RGB565
+        break;
+
+    case PIXFORMAT_YUV422:
+        write_reg(sensor->slv_addr, 0xfe, 0x00);
+        ret = set_reg_bits(sensor->slv_addr, 0x44, 0, 0x1f, 3);
+        break;
+    default:
+        ESP_LOGW(TAG, "unsupport format");
+        ret = -1;
+        break;
+    }
+    if (ret == 0) {
+        sensor->pixformat = pixformat;
+        ESP_LOGD(TAG, "Set pixformat to: %u", pixformat);
+    }
+
+    return ret;
+}
+
+static int set_framesize(sensor_t *sensor, framesize_t framesize)
+{
+    ESP_LOGI(TAG, "set_framesize");
+    int ret = 0;
+    if (framesize > FRAMESIZE_VGA) {
+        ESP_LOGW(TAG, "Invalid framesize: %u", framesize);
+        framesize = FRAMESIZE_VGA;
+    }
+    sensor->status.framesize = framesize;
+    uint16_t w = resolution[framesize].width;
+    uint16_t h = resolution[framesize].height;
+    uint16_t row_s = (resolution[FRAMESIZE_VGA].height - h) / 2;
+    uint16_t col_s = (resolution[FRAMESIZE_VGA].width - w) / 2;
+
+    write_reg(sensor->slv_addr, 0xfe, 0x00);
+    write_reg(sensor->slv_addr, P0_ROW_START_HIGH, H8(row_s)); // Row_start[8]
+    write_reg(sensor->slv_addr, P0_ROW_START_LOW, L8(row_s)); // Row_start[7:0]
+    write_reg(sensor->slv_addr, P0_COLUMN_START_HIGH, H8(col_s)); // Column_start[9:8]
+    write_reg(sensor->slv_addr, P0_COLUMN_START_LOW, L8(col_s)); // Column_start[7:0]
+    write_reg(sensor->slv_addr, P0_WINDOW_HEIGHT_HIGH, H8(h + 8)); //window_height [8]
+    write_reg(sensor->slv_addr, P0_WINDOW_HEIGHT_LOW, L8(h + 8)); //window_height [7:0]
+    write_reg(sensor->slv_addr, P0_WINDOW_WIDTH_HIGH, H8(w + 8)); //window_width [9:8]
+    write_reg(sensor->slv_addr, P0_WINDOW_WIDTH_LOW, L8(w + 8)); //window_width [7:0]
+
+    write_reg(sensor->slv_addr, P0_WIN_MODE, 0x01);
+    write_reg(sensor->slv_addr, P0_OUT_WIN_HEIGHT_HIGH, H8(h));
+    write_reg(sensor->slv_addr, P0_OUT_WIN_HEIGHT_LOW, L8(h));
+    write_reg(sensor->slv_addr, P0_OUT_WIN_WIDTH_HIGH, H8(w));
+    write_reg(sensor->slv_addr, P0_OUT_WIN_WIDTH_LOW, L8(w));
+
+    if (ret == 0) {
+        ESP_LOGD(TAG, "Set framesize to: %ux%u", w, h);
+    }
+    print_regs(sensor->slv_addr);
+    return ret;
+}
+
+static int set_hmirror(sensor_t *sensor, int enable)
+{
+    int ret = 0;
+    sensor->status.hmirror = enable;
+    ret = write_reg(sensor->slv_addr, 0xfe, 0x00);
+    ret |= set_reg_bits(sensor->slv_addr, P0_CISCTL_MODE1, 0, 0x01, enable);
+    if (ret == 0) {
+        ESP_LOGD(TAG, "Set h-mirror to: %d", enable);
+    }
+    return ret;
+}
+
+static int set_vflip(sensor_t *sensor, int enable)
+{
+    int ret = 0;
+    sensor->status.vflip = enable;
+    ret = write_reg(sensor->slv_addr, 0xfe, 0x00);
+    ret |= set_reg_bits(sensor->slv_addr, P0_CISCTL_MODE1, 1, 0x01, enable);
+    if (ret == 0) {
+        ESP_LOGD(TAG, "Set v-flip to: %d", enable);
+    }
+    return ret;
+}
+
+static int set_colorbar(sensor_t *sensor, int enable)
+{
+    int ret = 0;
+    ret = write_reg(sensor->slv_addr, 0xfe, 0x00);
+    ret |= set_reg_bits(sensor->slv_addr, P0_DEBUG_MODE2, 3, 0x01, enable);
+    if (ret == 0) {
+        sensor->status.colorbar = enable;
+        ESP_LOGD(TAG, "Set colorbar to: %d", enable);
+    }
+    return ret;
+}
+
+static int get_reg(sensor_t *sensor, int reg, int mask)
+{
+    int ret = 0;
+    if (mask > 0xFF) {
+        ESP_LOGE(TAG, "mask should not more than 0xff");
+    } else {
+        ret = read_reg(sensor->slv_addr, reg);
+    }
+    if (ret > 0) {
+        ret &= mask;
+    }
+    return ret;
+}
+
+static int set_reg(sensor_t *sensor, int reg, int mask, int value)
+{
+    int ret = 0;
+    if (mask > 0xFF) {
+        ESP_LOGE(TAG, "mask should not more than 0xff");
+    } else {
+        ret = read_reg(sensor->slv_addr, reg);
+    }
+    if (ret < 0) {
+        return ret;
+    }
+    value = (ret & ~mask) | (value & mask);
+
+    if (mask > 0xFF) {
+
+    } else {
+        ret = write_reg(sensor->slv_addr, reg, value);
+    }
+    return ret;
+}
+
+static int init_status(sensor_t *sensor)
+{
+    write_reg(sensor->slv_addr, 0xfe, 0x00);
+    sensor->status.brightness = 0;
+    sensor->status.contrast = 0;
+    sensor->status.saturation = 0;
+    sensor->status.sharpness = 0;
+    sensor->status.denoise = 0;
+    sensor->status.ae_level = 0;
+    sensor->status.gainceiling = 0;
+    sensor->status.awb = 0;
+    sensor->status.dcw = 0;
+    sensor->status.agc = 0;
+    sensor->status.aec = 0;
+    sensor->status.hmirror = check_reg_mask(sensor->slv_addr, P0_CISCTL_MODE1, 0x01);
+    sensor->status.vflip = check_reg_mask(sensor->slv_addr, P0_CISCTL_MODE1, 0x02);
+    sensor->status.colorbar = 0;
+    sensor->status.bpc = 0;
+    sensor->status.wpc = 0;
+    sensor->status.raw_gma = 0;
+    sensor->status.lenc = 0;
+    sensor->status.quality = 0;
+    sensor->status.special_effect = 0;
+    sensor->status.wb_mode = 0;
+    sensor->status.awb_gain = 0;
+    sensor->status.agc_gain = 0;
+    sensor->status.aec_value = 0;
+    sensor->status.aec2 = 0;
+    return 0;
+}
+
+static int set_dummy(sensor_t *sensor, int val)
+{
+    ESP_LOGW(TAG, "Unsupported");
+    return -1;
+}
+static int set_gainceiling_dummy(sensor_t *sensor, gainceiling_t val)
+{
+    ESP_LOGW(TAG, "Unsupported");
+    return -1;
+}
+
+int gc032a_detect(int slv_addr, sensor_id_t *id)
+{
+    if (GC032A_SCCB_ADDR == slv_addr) {
+        uint8_t MIDL = SCCB_Read(slv_addr, SENSOR_ID_LOW);
+        uint8_t MIDH = SCCB_Read(slv_addr, SENSOR_ID_HIGH);
+        uint16_t PID = MIDH << 8 | MIDL;
+        if (GC032A_PID == PID) {
+            id->PID = PID;
+            return PID;
+        } else {
+            ESP_LOGI(TAG, "Mismatch PID=0x%x", PID);
+        }
+    }
+    return 0;
+}
+
+int gc032a_init(sensor_t *sensor)
+{
+    sensor->init_status = init_status;
+    sensor->reset = reset;
+    sensor->set_pixformat = set_pixformat;
+    sensor->set_framesize = set_framesize;
+    sensor->set_contrast = set_dummy;
+    sensor->set_brightness = set_dummy;
+    sensor->set_saturation = set_dummy;
+    sensor->set_sharpness = set_dummy;
+    sensor->set_denoise = set_dummy;
+    sensor->set_gainceiling = set_gainceiling_dummy;
+    sensor->set_quality = set_dummy;
+    sensor->set_colorbar = set_colorbar;
+    sensor->set_whitebal = set_dummy;
+    sensor->set_gain_ctrl = set_dummy;
+    sensor->set_exposure_ctrl = set_dummy;
+    sensor->set_hmirror = set_hmirror;
+    sensor->set_vflip = set_vflip;
+
+    sensor->set_aec2 = set_dummy;
+    sensor->set_awb_gain = set_dummy;
+    sensor->set_agc_gain = set_dummy;
+    sensor->set_aec_value = set_dummy;
+
+    sensor->set_special_effect = set_dummy;
+    sensor->set_wb_mode = set_dummy;
+    sensor->set_ae_level = set_dummy;
+
+    sensor->set_dcw = set_dummy;
+    sensor->set_bpc = set_dummy;
+    sensor->set_wpc = set_dummy;
+
+    sensor->set_raw_gma = set_dummy;
+    sensor->set_lenc = set_dummy;
+
+    sensor->get_reg = get_reg;
+    sensor->set_reg = set_reg;
+    sensor->set_res_raw = NULL;
+    sensor->set_pll = NULL;
+    sensor->set_xclk = NULL;
+
+    ESP_LOGD(TAG, "GC032A Attached");
+    return 0;
+}
--- a/code/components/esp32-camera-master/sensors/gc2145.c
+++ b/code/components/esp32-camera-master/sensors/gc2145.c
@@ -0,0 +1,477 @@
+// Copyright 2015-2021 Espressif Systems (Shanghai) PTE LTD
+//
+// Licensed under the Apache License, Version 2.0 (the "License");
+// you may not use this file except in compliance with the License.
+// You may obtain a copy of the License at
+
+//     http://www.apache.org/licenses/LICENSE-2.0
+//
+// Unless required by applicable law or agreed to in writing, software
+// distributed under the License is distributed on an "AS IS" BASIS,
+// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+// See the License for the specific language governing permissions and
+// limitations under the License.
+
+#include <stdint.h>
+#include <stdlib.h>
+#include <string.h>
+#include "freertos/FreeRTOS.h"
+#include "freertos/task.h"
+#include "sccb.h"
+#include "gc2145.h"
+#include "gc2145_regs.h"
+#include "gc2145_settings.h"
+
+#if defined(ARDUINO_ARCH_ESP32) && defined(CONFIG_ARDUHAL_ESP_LOG)
+#include "esp32-hal-log.h"
+#else
+#include "esp_log.h"
+static const char *TAG = "gc2145";
+#endif
+
+#define H8(v) ((v)>>8)
+#define L8(v) ((v)&0xff)
+
+//#define REG_DEBUG_ON
+
+static int read_reg(uint8_t slv_addr, const uint16_t reg)
+{
+    int ret = SCCB_Read(slv_addr, reg);
+#ifdef REG_DEBUG_ON
+    if (ret < 0) {
+        ESP_LOGE(TAG, "READ REG 0x%04x FAILED: %d", reg, ret);
+    }
+#endif
+    return ret;
+}
+
+static int write_reg(uint8_t slv_addr, const uint16_t reg, uint8_t value)
+{
+    int ret = 0;
+#ifndef REG_DEBUG_ON
+    ret = SCCB_Write(slv_addr, reg, value);
+#else
+    int old_value = read_reg(slv_addr, reg);
+    if (old_value < 0) {
+        return old_value;
+    }
+    if ((uint8_t)old_value != value) {
+        ESP_LOGI(TAG, "NEW REG 0x%04x: 0x%02x to 0x%02x", reg, (uint8_t)old_value, value);
+        ret = SCCB_Write(slv_addr, reg, value);
+    } else {
+        ESP_LOGD(TAG, "OLD REG 0x%04x: 0x%02x", reg, (uint8_t)old_value);
+        ret = SCCB_Write(slv_addr, reg, value);//maybe not?
+    }
+    if (ret < 0) {
+        ESP_LOGE(TAG, "WRITE REG 0x%04x FAILED: %d", reg, ret);
+    }
+#endif
+    return ret;
+}
+
+static int check_reg_mask(uint8_t slv_addr, uint16_t reg, uint8_t mask)
+{
+    return (read_reg(slv_addr, reg) & mask) == mask;
+}
+
+static int set_reg_bits(uint8_t slv_addr, uint16_t reg, uint8_t offset, uint8_t mask, uint8_t value)
+{
+    int ret = 0;
+    uint8_t c_value, new_value;
+    ret = read_reg(slv_addr, reg);
+    if (ret < 0) {
+        return ret;
+    }
+    c_value = ret;
+    new_value = (c_value & ~(mask << offset)) | ((value & mask) << offset);
+    ret = write_reg(slv_addr, reg, new_value);
+    return ret;
+}
+
+static int write_regs(uint8_t slv_addr, const uint16_t (*regs)[2])
+{
+    int i = 0, ret = 0;
+    while (!ret && regs[i][0] != REGLIST_TAIL) {
+        if (regs[i][0] == REG_DLY) {
+            vTaskDelay(regs[i][1] / portTICK_PERIOD_MS);
+        } else {
+            ret = write_reg(slv_addr, regs[i][0], regs[i][1]);
+        }
+        i++;
+    }
+    return ret;
+}
+
+static void print_regs(uint8_t slv_addr)
+{
+#ifdef DEBUG_PRINT_REG
+    vTaskDelay(pdMS_TO_TICKS(100));
+    ESP_LOGI(TAG, "REG list look ======================");
+    for (size_t i = 0xf0; i <= 0xfe; i++) {
+        ESP_LOGI(TAG, "reg[0x%02x] = 0x%02x", i, read_reg(slv_addr, i));
+    }
+    ESP_LOGI(TAG, "\npage 0 ===");
+    write_reg(slv_addr, 0xfe, 0x00); // page 0
+    for (size_t i = 0x03; i <= 0x24; i++) {
+        ESP_LOGI(TAG, "p0 reg[0x%02x] = 0x%02x", i, read_reg(slv_addr, i));
+    }
+    for (size_t i = 0x80; i <= 0xa2; i++) {
+        ESP_LOGI(TAG, "p0 reg[0x%02x] = 0x%02x", i, read_reg(slv_addr, i));
+    }
+    ESP_LOGI(TAG, "\npage 3 ===");
+    write_reg(slv_addr, 0xfe, 0x03); // page 3
+    for (size_t i = 0x01; i <= 0x43; i++) {
+        ESP_LOGI(TAG, "p3 reg[0x%02x] = 0x%02x", i, read_reg(slv_addr, i));
+    }
+#endif
+}
+
+static int reset(sensor_t *sensor)
+{
+    int ret = 0;
+    // Software Reset: clear all registers and reset them to their default values
+    ret = write_reg(sensor->slv_addr, RESET_RELATED, 0xe0);
+    if (ret) {
+        ESP_LOGE(TAG, "Software Reset FAILED!");
+        return ret;
+    }
+    vTaskDelay(100 / portTICK_PERIOD_MS);
+    ret = write_regs(sensor->slv_addr, gc2145_default_init_regs);
+    if (ret == 0) {
+        ESP_LOGD(TAG, "Camera defaults loaded");
+        vTaskDelay(100 / portTICK_PERIOD_MS);
+#ifdef CONFIG_IDF_TARGET_ESP32
+        write_reg(sensor->slv_addr, 0xfe, 0x00);
+        //ensure pclk <= 15MHz for esp32
+        set_reg_bits(sensor->slv_addr, 0xf8, 0, 0x3f, 2); // divx4
+        set_reg_bits(sensor->slv_addr, 0xfa, 4, 0x0f, 2); // divide_by
+#endif
+
+    }
+    return ret;
+}
+
+static int set_pixformat(sensor_t *sensor, pixformat_t pixformat)
+{
+    int ret = 0;
+
+    switch (pixformat) {
+    case PIXFORMAT_RGB565:
+        write_reg(sensor->slv_addr, 0xfe, 0x00);
+        ret = set_reg_bits(sensor->slv_addr, P0_OUTPUT_FORMAT, 0, 0x1f, 6);  //RGB565
+        break;
+
+    case PIXFORMAT_YUV422:
+        write_reg(sensor->slv_addr, 0xfe, 0x00);
+        ret = set_reg_bits(sensor->slv_addr, P0_OUTPUT_FORMAT, 0, 0x1f, 2); //yuv422
+        break;
+    default:
+        ESP_LOGW(TAG, "unsupport format");
+        ret = -1;
+        break;
+    }
+
+    if (ret == 0) {
+        sensor->pixformat = pixformat;
+        ESP_LOGD(TAG, "Set pixformat to: %u", pixformat);
+    }
+    return ret;
+}
+
+static int set_framesize(sensor_t *sensor, framesize_t framesize)
+{
+    int ret = 0;
+    if (framesize > FRAMESIZE_UXGA) {
+        ESP_LOGW(TAG, "Invalid framesize: %u", framesize);
+        framesize = FRAMESIZE_UXGA;
+    }
+    sensor->status.framesize = framesize;
+    uint16_t w = resolution[framesize].width;
+    uint16_t h = resolution[framesize].height;
+    uint16_t row_s = (resolution[FRAMESIZE_UXGA].height - h) / 2;
+    uint16_t col_s = (resolution[FRAMESIZE_UXGA].width - w) / 2;
+    (void)row_s;
+    (void)col_s;
+
+#if CONFIG_GC_SENSOR_SUBSAMPLE_MODE
+    struct subsample_cfg {
+        uint16_t ratio_numerator;
+        uint16_t ratio_denominator;
+        uint8_t reg0x99;
+        uint8_t reg0x9b;
+        uint8_t reg0x9c;
+        uint8_t reg0x9d;
+        uint8_t reg0x9e;
+        uint8_t reg0x9f;
+        uint8_t reg0xa0;
+        uint8_t reg0xa1;
+        uint8_t reg0xa2;
+    };
+    const struct subsample_cfg subsample_cfgs[] = { // define some subsample ratio
+        // {60, 420, 0x77, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00}, //1/7 // A smaller ratio brings a larger view, but it reduces the frame rate
+        // {84, 420, 0x55, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00}, //1/5
+        // {105, 420, 0x44, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00},//1/4
+        {140, 420, 0x33, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00},//1/3
+        {210, 420, 0x22, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00},//1/2
+        {240, 420, 0x77, 0x02, 0x46, 0x02, 0x46, 0x02, 0x46, 0x02, 0x46},//4/7
+        {252, 420, 0x55, 0x02, 0x04, 0x02, 0x04, 0x02, 0x04, 0x02, 0x04},//3/5
+        {280, 420, 0x33, 0x00, 0x02, 0x00, 0x00, 0x00, 0x02, 0x00, 0x00},//2/3
+        {420, 420, 0x11, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00, 0x00},//1/1
+    };
+    uint16_t win_w = resolution[FRAMESIZE_UXGA].width;
+    uint16_t win_h = resolution[FRAMESIZE_UXGA].height;
+    const struct subsample_cfg *cfg = NULL;
+    /**
+     * Strategy: try to keep the maximum perspective
+     */
+    uint8_t i = 0;
+    if (framesize >= FRAMESIZE_QVGA) {
+        i = 1;
+    }
+    for (; i < sizeof(subsample_cfgs) / sizeof(struct subsample_cfg); i++) {
+        cfg = &subsample_cfgs[i];
+        if ((win_w * cfg->ratio_numerator / cfg->ratio_denominator >= w) && (win_h * cfg->ratio_numerator / cfg->ratio_denominator >= h)) {
+            win_w = w * cfg->ratio_denominator / cfg->ratio_numerator;
+            win_h = h * cfg->ratio_denominator / cfg->ratio_numerator;
+            row_s = (resolution[FRAMESIZE_UXGA].height - win_h) / 2;
+            col_s = (resolution[FRAMESIZE_UXGA].width - win_w) / 2;
+            ESP_LOGI(TAG, "subsample win:%dx%d, ratio:%f", win_w, win_h, (float)cfg->ratio_numerator / (float)cfg->ratio_denominator);
+            break;
+        }
+    }
+
+    write_reg(sensor->slv_addr, 0xfe, 0x00);
+    write_reg(sensor->slv_addr, P0_CROP_ENABLE, 0x01);
+    write_reg(sensor->slv_addr, 0x09, H8(row_s));
+    write_reg(sensor->slv_addr, 0x0a, L8(row_s));
+    write_reg(sensor->slv_addr, 0x0b, H8(col_s));
+    write_reg(sensor->slv_addr, 0x0c, L8(col_s));
+    write_reg(sensor->slv_addr, 0x0d, H8(win_h + 8));
+    write_reg(sensor->slv_addr, 0x0e, L8(win_h + 8));
+    write_reg(sensor->slv_addr, 0x0f, H8(win_w + 16));
+    write_reg(sensor->slv_addr, 0x10, L8(win_w + 16));
+
+    write_reg(sensor->slv_addr, 0x99, cfg->reg0x99);
+    write_reg(sensor->slv_addr, 0x9b, cfg->reg0x9b);
+    write_reg(sensor->slv_addr, 0x9c, cfg->reg0x9c);
+    write_reg(sensor->slv_addr, 0x9d, cfg->reg0x9d);
+    write_reg(sensor->slv_addr, 0x9e, cfg->reg0x9e);
+    write_reg(sensor->slv_addr, 0x9f, cfg->reg0x9f);
+    write_reg(sensor->slv_addr, 0xa0, cfg->reg0xa0);
+    write_reg(sensor->slv_addr, 0xa1, cfg->reg0xa1);
+    write_reg(sensor->slv_addr, 0xa2, cfg->reg0xa2);
+
+    write_reg(sensor->slv_addr, 0x95, H8(h));
+    write_reg(sensor->slv_addr, 0x96, L8(h));
+    write_reg(sensor->slv_addr, 0x97, H8(w));
+    write_reg(sensor->slv_addr, 0x98, L8(w));
+
+
+#elif CONFIG_GC_SENSOR_WINDOWING_MODE
+    write_reg(sensor->slv_addr, 0xfe, 0x00);
+
+    write_reg(sensor->slv_addr, P0_CROP_ENABLE, 0x01);
+    // write_reg(sensor->slv_addr, 0xec, col_s / 8); //measure window
+    // write_reg(sensor->slv_addr, 0xed, row_s / 8);
+    // write_reg(sensor->slv_addr, 0xee, (col_s + h) / 8);
+    // write_reg(sensor->slv_addr, 0xef, (row_s + w) / 8);
+
+    write_reg(sensor->slv_addr, 0x09, H8(row_s));
+    write_reg(sensor->slv_addr, 0x0a, L8(row_s));
+    write_reg(sensor->slv_addr, 0x0b, H8(col_s));
+    write_reg(sensor->slv_addr, 0x0c, L8(col_s));
+    write_reg(sensor->slv_addr, 0x0d, H8(h + 8));
+    write_reg(sensor->slv_addr, 0x0e, L8(h + 8));
+    write_reg(sensor->slv_addr, 0x0f, H8(w + 8));
+    write_reg(sensor->slv_addr, 0x10, L8(w + 8));
+
+    write_reg(sensor->slv_addr, 0x95, H8(h));
+    write_reg(sensor->slv_addr, 0x96, L8(h));
+    write_reg(sensor->slv_addr, 0x97, H8(w));
+    write_reg(sensor->slv_addr, 0x98, L8(w));
+
+#endif
+
+    if (ret == 0) {
+        ESP_LOGD(TAG, "Set framesize to: %ux%u", w, h);
+    }
+    return ret;
+
+}
+
+static int set_hmirror(sensor_t *sensor, int enable)
+{
+    int ret = 0;
+    sensor->status.hmirror = enable;
+    ret = write_reg(sensor->slv_addr, 0xfe, 0x00);
+    ret |= set_reg_bits(sensor->slv_addr, P0_ANALOG_MODE1, 0, 0x01, enable != 0);
+    if (ret == 0) {
+        ESP_LOGD(TAG, "Set h-mirror to: %d", enable);
+    }
+    return ret;
+}
+
+static int set_vflip(sensor_t *sensor, int enable)
+{
+    int ret = 0;
+    sensor->status.vflip = enable;
+    ret = write_reg(sensor->slv_addr, 0xfe, 0x00);
+    ret |= set_reg_bits(sensor->slv_addr, P0_ANALOG_MODE1, 1, 0x01, enable != 0);
+    if (ret == 0) {
+        ESP_LOGD(TAG, "Set v-flip to: %d", enable);
+    }
+    return ret;
+}
+
+static int set_colorbar(sensor_t *sensor, int enable)
+{
+    int ret = 0;
+    // ret = write_reg(sensor->slv_addr, 0xfe, 0x00);
+    // ret |= set_reg_bits(sensor->slv_addr, P0_DEBUG_MODE3, 3, 0x01, enable);
+    if (ret == 0) {
+        sensor->status.colorbar = enable;
+        ESP_LOGD(TAG, "Set colorbar to: %d", enable);
+    }
+    return ret;
+}
+
+static int get_reg(sensor_t *sensor, int reg, int mask)
+{
+    int ret = 0;
+    if (mask > 0xFF) {
+        ESP_LOGE(TAG, "mask should not more than 0xff");
+    } else {
+        ret = read_reg(sensor->slv_addr, reg);
+    }
+    if (ret > 0) {
+        ret &= mask;
+    }
+    return ret;
+}
+
+static int set_reg(sensor_t *sensor, int reg, int mask, int value)
+{
+    int ret = 0;
+    if (mask > 0xFF) {
+        ESP_LOGE(TAG, "mask should not more than 0xff");
+    } else {
+        ret = read_reg(sensor->slv_addr, reg);
+    }
+    if (ret < 0) {
+        return ret;
+    }
+    value = (ret & ~mask) | (value & mask);
+
+    if (mask > 0xFF) {
+
+    } else {
+        ret = write_reg(sensor->slv_addr, reg, value);
+    }
+    return ret;
+}
+
+static int init_status(sensor_t *sensor)
+{
+    write_reg(sensor->slv_addr, 0xfe, 0x00);
+    sensor->status.brightness = 0;
+    sensor->status.contrast = 0;
+    sensor->status.saturation = 0;
+    sensor->status.sharpness = 0;
+    sensor->status.denoise = 0;
+    sensor->status.ae_level = 0;
+    sensor->status.gainceiling = 0;
+    sensor->status.awb = 0;
+    sensor->status.dcw = 0;
+    sensor->status.agc = 0;
+    sensor->status.aec = 0;
+    sensor->status.hmirror = check_reg_mask(sensor->slv_addr, P0_ANALOG_MODE1, 0x01);
+    sensor->status.vflip = check_reg_mask(sensor->slv_addr, P0_ANALOG_MODE1, 0x02);
+    sensor->status.colorbar = 0;
+    sensor->status.bpc = 0;
+    sensor->status.wpc = 0;
+    sensor->status.raw_gma = 0;
+    sensor->status.lenc = 0;
+    sensor->status.quality = 0;
+    sensor->status.special_effect = 0;
+    sensor->status.wb_mode = 0;
+    sensor->status.awb_gain = 0;
+    sensor->status.agc_gain = 0;
+    sensor->status.aec_value = 0;
+    sensor->status.aec2 = 0;
+
+    print_regs(sensor->slv_addr);
+    return 0;
+}
+
+static int set_dummy(sensor_t *sensor, int val)
+{
+    ESP_LOGW(TAG, "Unsupported");
+    return -1;
+}
+static int set_gainceiling_dummy(sensor_t *sensor, gainceiling_t val)
+{
+    ESP_LOGW(TAG, "Unsupported");
+    return -1;
+}
+
+int gc2145_detect(int slv_addr, sensor_id_t *id)
+{
+    if (GC2145_SCCB_ADDR == slv_addr) {
+        uint8_t MIDL = SCCB_Read(slv_addr, CHIP_ID_LOW);
+        uint8_t MIDH = SCCB_Read(slv_addr, CHIP_ID_HIGH);
+        uint16_t PID = MIDH << 8 | MIDL;
+        if (GC2145_PID == PID) {
+            id->PID = PID;
+            return PID;
+        } else {
+            ESP_LOGI(TAG, "Mismatch PID=0x%x", PID);
+        }
+    }
+    return 0;
+}
+
+int gc2145_init(sensor_t *sensor)
+{
+    sensor->init_status = init_status;
+    sensor->reset = reset;
+    sensor->set_pixformat = set_pixformat;
+    sensor->set_framesize = set_framesize;
+    sensor->set_contrast = set_dummy;
+    sensor->set_brightness = set_dummy;
+    sensor->set_saturation = set_dummy;
+    sensor->set_sharpness = set_dummy;
+    sensor->set_denoise = set_dummy;
+    sensor->set_gainceiling = set_gainceiling_dummy;
+    sensor->set_quality = set_dummy;
+    sensor->set_colorbar = set_colorbar;
+    sensor->set_whitebal = set_dummy;
+    sensor->set_gain_ctrl = set_dummy;
+    sensor->set_exposure_ctrl = set_dummy;
+    sensor->set_hmirror = set_hmirror;
+    sensor->set_vflip = set_vflip;
+
+    sensor->set_aec2 = set_dummy;
+    sensor->set_awb_gain = set_dummy;
+    sensor->set_agc_gain = set_dummy;
+    sensor->set_aec_value = set_dummy;
+
+    sensor->set_special_effect = set_dummy;
+    sensor->set_wb_mode = set_dummy;
+    sensor->set_ae_level = set_dummy;
+
+    sensor->set_dcw = set_dummy;
+    sensor->set_bpc = set_dummy;
+    sensor->set_wpc = set_dummy;
+
+    sensor->set_raw_gma = set_dummy;
+    sensor->set_lenc = set_dummy;
+
+    sensor->get_reg = get_reg;
+    sensor->set_reg = set_reg;
+    sensor->set_res_raw = NULL;
+    sensor->set_pll = NULL;
+    sensor->set_xclk = NULL;
+
+    ESP_LOGD(TAG, "GC2145 Attached");
+    return 0;
+}
--- a/code/components/esp32-camera-master/sensors/nt99141.c
+++ b/code/components/esp32-camera-master/sensors/nt99141.c
@@ -10,6 +10,7 @@
 #include <stdlib.h>
 #include <string.h>
 #include "sccb.h"
+#include "xclk.h"
 #include "nt99141.h"
 #include "nt99141_regs.h"
 #include "nt99141_settings.h"
@@ -144,28 +145,6 @@ static int write_addr_reg(uint8_t slv_addr, const uint16_t reg, uint16_t x_value

 #define write_reg_bits(slv_addr, reg, mask, enable) set_reg_bits(slv_addr, reg, 0, mask, enable?mask:0)

-static int calc_sysclk(int xclk, bool pll_bypass, int pll_multiplier, int pll_sys_div, int pll_pre_div, bool pll_root_2x, int pll_seld5, bool pclk_manual, int pclk_div)
-{
-    const int pll_pre_div2x_map[] = { 2, 3, 4, 6 };//values are multiplied by two to avoid floats
-    const int pll_seld52x_map[] = { 2, 2, 4, 5 };
-
-    if (!pll_sys_div) {
-        pll_sys_div = 1;
-    }
-
-    int pll_pre_div2x = pll_pre_div2x_map[pll_pre_div];
-    int pll_root_div = pll_root_2x ? 2 : 1;
-    int pll_seld52x = pll_seld52x_map[pll_seld5];
-
-    int VCO = (xclk / 1000) * pll_multiplier * pll_root_div * 2 / pll_pre_div2x;
-    int PLLCLK = pll_bypass ? (xclk) : (VCO * 1000 * 2 / pll_sys_div / pll_seld52x);
-    int PCLK = PLLCLK / 2 / ((pclk_manual && pclk_div) ? pclk_div : 1);
-    int SYSCLK = PLLCLK / 4;
-
-    ESP_LOGD(TAG, "Calculated VCO: %d Hz, PLLCLK: %d Hz, SYSCLK: %d Hz, PCLK: %d Hz", VCO * 1000, PLLCLK, SYSCLK, PCLK);
-    return SYSCLK;
-}
-
 static int set_pll(sensor_t *sensor, bool bypass, uint8_t multiplier, uint8_t sys_div, uint8_t pre_div, bool root_2x, uint8_t seld5, bool pclk_manual, uint8_t pclk_div)
 {
    return -1;
@@ -309,7 +288,7 @@ static int set_framesize(sensor_t *sensor, framesize_t framesize)
        ret = write_regs(sensor->slv_addr, sensor_framesize_VGA);
    }

-    return 0;
+    return ret;
 }

 static int set_hmirror(sensor_t *sensor, int enable)
@@ -682,7 +661,6 @@ static int set_brightness(sensor_t *sensor, int level)
 {
    int ret = 0;
    uint8_t value = 0;
-    bool negative = false;

    switch (level) {
        case 3:
@@ -699,17 +677,14 @@ static int set_brightness(sensor_t *sensor, int level)

        case -1:
            value = 0x78;
-            negative = true;
            break;

        case -2:
            value = 0x70;
-            negative = true;
            break;

        case -3:
            value = 0x60;
-            negative = true;
            break;

        default: // 0
@@ -730,7 +705,6 @@ static int set_contrast(sensor_t *sensor, int level)
 {
    int ret = 0;
    uint8_t value1 = 0, value2 = 0 ;
-    bool negative = false;

    switch (level) {
        case 3:
@@ -947,7 +921,6 @@ static int _set_pll(sensor_t *sensor, int bypass, int multiplier, int sys_div, i
    return set_pll(sensor, bypass > 0, multiplier, sys_div, pre_div, root_2x > 0, seld5, pclk_manual > 0, pclk_div);
 }

-esp_err_t xclk_timer_conf(int ledc_timer, int xclk_freq_hz);
 static int set_xclk(sensor_t *sensor, int timer, int xclk)
 {
    int ret = 0;
@@ -961,6 +934,23 @@ static int set_xclk(sensor_t *sensor, int timer, int xclk)
    return ret;
 }

+int nt99141_detect(int slv_addr, sensor_id_t *id)
+{
+    if (NT99141_SCCB_ADDR == slv_addr) {
+        SCCB_Write16(slv_addr, 0x3008, 0x01);//bank sensor
+        uint16_t h = SCCB_Read16(slv_addr, 0x3000);
+        uint16_t l = SCCB_Read16(slv_addr, 0x3001);
+        uint16_t PID = (h<<8) | l;
+        if (NT99141_PID == PID) {
+            id->PID = PID;
+            return PID;
+        } else {
+            ESP_LOGI(TAG, "Mismatch PID=0x%x", PID);
+        }
+    }
+    return 0;
+}
+
 static int init_status(sensor_t *sensor)
 {
    sensor->status.brightness = 0;
@@ -991,7 +981,7 @@ static int init_status(sensor_t *sensor)
    return 0;
 }

-int NT99141_init(sensor_t *sensor)
+int nt99141_init(sensor_t *sensor)
 {
    sensor->reset = reset;
    sensor->set_pixformat = set_pixformat;
--- a/code/components/esp32-camera-master/sensors/ov2640.c
+++ b/code/components/esp32-camera-master/sensors/ov2640.c
@@ -10,6 +10,7 @@
 #include <stdlib.h>
 #include <string.h>
 #include "sccb.h"
+#include "xclk.h"
 #include "ov2640.h"
 #include "ov2640_regs.h"
 #include "ov2640_settings.h"
@@ -149,7 +150,7 @@ static int set_window(sensor_t *sensor, ov2640_sensor_mode_t mode, int offset_x,
        {VSIZE, max_y & 0xFF},
        {XOFFL, offset_x & 0xFF},
        {YOFFL, offset_y & 0xFF},
-        {VHYX, ((max_y >> 1) & 0X80) | ((offset_y >> 4) & 0X70) | ((max_x >> 5) & 0X08) | ((offset_y >> 8) & 0X07)},
+        {VHYX, ((max_y >> 1) & 0X80) | ((offset_y >> 4) & 0X70) | ((max_x >> 5) & 0X08) | ((offset_x >> 8) & 0X07)},
        {TEST, (max_x >> 2) & 0X80},
        {ZMOW, (w)&0xFF},
        {ZMOH, (h)&0xFF},
@@ -157,26 +158,40 @@ static int set_window(sensor_t *sensor, ov2640_sensor_mode_t mode, int offset_x,
        {0, 0}
    };

-    c.pclk_auto = 0;
-    c.pclk_div = 8;
-    c.clk_2x = 0;
-    c.clk_div = 0;
-
-    if(sensor->pixformat != PIXFORMAT_JPEG){
-        c.pclk_auto = 1;
+    if (sensor->pixformat == PIXFORMAT_JPEG) {
+        c.clk_2x = 0;
+        c.clk_div = 0;
+        c.pclk_auto = 0;
+        c.pclk_div = 8;
+        if(mode == OV2640_MODE_UXGA) {
+            c.pclk_div = 12;
+        }
+        // if (sensor->xclk_freq_hz == 16000000) {
+        //     c.pclk_div = c.pclk_div / 2;
+        // }
+    } else {
+#if CONFIG_IDF_TARGET_ESP32
+        c.clk_2x = 0;
+#else
+        c.clk_2x = 1;
+#endif
        c.clk_div = 7;
+        c.pclk_auto = 1;
+        c.pclk_div = 8;
+        if (mode == OV2640_MODE_CIF) {
+            c.clk_div = 3;
+        } else if(mode == OV2640_MODE_UXGA) {
+            c.pclk_div = 12;
+        }
    }
+    ESP_LOGI(TAG, "Set PLL: clk_2x: %u, clk_div: %u, pclk_auto: %u, pclk_div: %u", c.clk_2x, c.clk_div, c.pclk_auto, c.pclk_div);

    if (mode == OV2640_MODE_CIF) {
        regs = ov2640_settings_to_cif;
-        if(sensor->pixformat != PIXFORMAT_JPEG){
-            c.clk_div = 3;
-        }
    } else if (mode == OV2640_MODE_SVGA) {
        regs = ov2640_settings_to_svga;
    } else {
        regs = ov2640_settings_to_uxga;
-        c.pclk_div = 12;
    }

    WRITE_REG_OR_RETURN(BANK_DSP, R_BYPASS, R_BYPASS_DSP_BYPAS);
@@ -480,7 +495,6 @@ static int _set_pll(sensor_t *sensor, int bypass, int multiplier, int sys_div, i
    return -1;
 }

-esp_err_t xclk_timer_conf(int ledc_timer, int xclk_freq_hz);
 static int set_xclk(sensor_t *sensor, int timer, int xclk)
 {
    int ret = 0;
@@ -531,6 +545,24 @@ static int init_status(sensor_t *sensor){
    return 0;
 }

+int ov2640_detect(int slv_addr, sensor_id_t *id)
+{
+    if (OV2640_SCCB_ADDR == slv_addr) {
+        SCCB_Write(slv_addr, 0xFF, 0x01);//bank sensor
+        uint16_t PID = SCCB_Read(slv_addr, 0x0A);
+        if (OV2640_PID == PID) {
+            id->PID = PID;
+            id->VER = SCCB_Read(slv_addr, REG_VER);
+            id->MIDL = SCCB_Read(slv_addr, REG_MIDL);
+            id->MIDH = SCCB_Read(slv_addr, REG_MIDH);
+            return PID;
+        } else {
+            ESP_LOGI(TAG, "Mismatch PID=0x%x", PID);
+        }
+    }
+    return 0;
+}
+
 int ov2640_init(sensor_t *sensor)
 {
    sensor->reset = reset;
--- a/code/components/esp32-camera-master/sensors/ov3660.c
+++ b/code/components/esp32-camera-master/sensors/ov3660.c
@@ -10,6 +10,7 @@
 #include <stdlib.h>
 #include <string.h>
 #include "sccb.h"
+#include "xclk.h"
 #include "ov3660.h"
 #include "ov3660_regs.h"
 #include "ov3660_settings.h"
@@ -142,7 +143,7 @@ static int calc_sysclk(int xclk, bool pll_bypass, int pll_multiplier, int pll_sy
    int PCLK = PLLCLK / 2 / ((pclk_manual && pclk_div)?pclk_div:1);
    int SYSCLK = PLLCLK / 4;

-    ESP_LOGD(TAG, "Calculated VCO: %d Hz, PLLCLK: %d Hz, SYSCLK: %d Hz, PCLK: %d Hz", VCO*1000, PLLCLK, SYSCLK, PCLK);
+    ESP_LOGI(TAG, "Calculated VCO: %d Hz, PLLCLK: %d Hz, SYSCLK: %d Hz, PCLK: %d Hz", VCO*1000, PLLCLK, SYSCLK, PCLK);
    return SYSCLK;
 }

@@ -310,13 +311,13 @@ static int set_image_options(sensor_t *sensor)
 static int set_framesize(sensor_t *sensor, framesize_t framesize)
 {
    int ret = 0;
-    framesize_t old_framesize = sensor->status.framesize;
-    sensor->status.framesize = framesize;

    if(framesize > FRAMESIZE_QXGA){
-        ESP_LOGE(TAG, "Invalid framesize: %u", framesize);
-        return -1;
+        ESP_LOGW(TAG, "Invalid framesize: %u", framesize);
+        framesize = FRAMESIZE_QXGA;
    }
+    framesize_t old_framesize = sensor->status.framesize;
+    sensor->status.framesize = framesize;
    uint16_t w = resolution[framesize].width;
    uint16_t h = resolution[framesize].height;
    aspect_ratio_t ratio = resolution[sensor->status.framesize].aspect_ratio;
@@ -355,7 +356,7 @@ static int set_framesize(sensor_t *sensor, framesize_t framesize)
    }

    if (sensor->pixformat == PIXFORMAT_JPEG) {
-        if (framesize == FRAMESIZE_QXGA) {
+        if (framesize == FRAMESIZE_QXGA || sensor->xclk_freq_hz == 16000000) {
            //40MHz SYSCLK and 10MHz PCLK
            ret = set_pll(sensor, false, 24, 1, 3, false, 0, true, 8);
        } else {
@@ -363,12 +364,16 @@ static int set_framesize(sensor_t *sensor, framesize_t framesize)
            ret = set_pll(sensor, false, 30, 1, 3, false, 0, true, 10);
        }
    } else {
-        if (framesize > FRAMESIZE_CIF) {
-            //10MHz SYSCLK and 10MHz PCLK (6.19 FPS)
-            ret = set_pll(sensor, false, 2, 1, 0, false, 0, true, 2);
+        //tuned for 16MHz XCLK and 8MHz PCLK
+        if (framesize > FRAMESIZE_HVGA) {
+            //8MHz SYSCLK and 8MHz PCLK (4.44 FPS)
+            ret = set_pll(sensor, false, 4, 1, 0, false, 2, true, 2);
+        } else if (framesize >= FRAMESIZE_QVGA) {
+            //16MHz SYSCLK and 8MHz PCLK (10.25 FPS)
+            ret = set_pll(sensor, false, 8, 1, 0, false, 2, true, 4);
        } else {
-            //25MHz SYSCLK and 10MHz PCLK (15.45 FPS)
-            ret = set_pll(sensor, false, 5, 1, 0, false, 0, true, 5);
+            //32MHz SYSCLK and 8MHz PCLK (17.77 FPS)
+            ret = set_pll(sensor, false, 8, 1, 0, false, 0, true, 8);
        }
    }

@@ -953,7 +958,6 @@ static int _set_pll(sensor_t *sensor, int bypass, int multiplier, int sys_div, i
    return set_pll(sensor, bypass > 0, multiplier, sys_div, pre_div, root_2x > 0, seld5, pclk_manual > 0, pclk_div);
 }

-esp_err_t xclk_timer_conf(int ledc_timer, int xclk_freq_hz);
 static int set_xclk(sensor_t *sensor, int timer, int xclk)
 {
    int ret = 0;
@@ -992,6 +996,22 @@ static int init_status(sensor_t *sensor)
    return 0;
 }

+int ov3660_detect(int slv_addr, sensor_id_t *id)
+{
+    if (OV3660_SCCB_ADDR == slv_addr) {
+        uint8_t h = SCCB_Read16(slv_addr, 0x300A);
+        uint8_t l = SCCB_Read16(slv_addr, 0x300B);
+        uint16_t PID = (h<<8) | l;
+        if (OV3660_PID == PID) {
+            id->PID = PID;
+            return PID;
+        } else {
+            ESP_LOGI(TAG, "Mismatch PID=0x%x", PID);
+        }
+    }
+    return 0;
+}
+
 int ov3660_init(sensor_t *sensor)
 {
    sensor->reset = reset;
--- a/code/components/esp32-camera-master/sensors/ov5640.c
+++ b/code/components/esp32-camera-master/sensors/ov5640.c
@@ -10,6 +10,7 @@
 #include <stdlib.h>
 #include <string.h>
 #include "sccb.h"
+#include "xclk.h"
 #include "ov5640.h"
 #include "ov5640_regs.h"
 #include "ov5640_settings.h"
@@ -196,7 +197,7 @@ static int calc_sysclk(int xclk, bool pll_bypass, int pll_multiplier, int pll_sy

    unsigned int SYSCLK = PLL_CLK / 4;

-    ESP_LOGD(TAG, "Calculated XVCLK: %d Hz, REFIN: %u Hz, VCO: %u Hz, PLL_CLK: %u Hz, SYSCLK: %u Hz, PCLK: %u Hz", xclk, REFIN, VCO, PLL_CLK, SYSCLK, PCLK);
+    ESP_LOGI(TAG, "Calculated XVCLK: %d Hz, REFIN: %u Hz, VCO: %u Hz, PLL_CLK: %u Hz, SYSCLK: %u Hz, PCLK: %u Hz", xclk, REFIN, VCO, PLL_CLK, SYSCLK, PCLK);
    return SYSCLK;
 }

@@ -209,6 +210,7 @@ static int set_pll(sensor_t *sensor, bool bypass, uint8_t multiplier, uint8_t sy
    if(multiplier > 127){
        multiplier &= 0xFE;//only even integers above 127
    }
+    ESP_LOGI(TAG, "Set PLL: bypass: %u, multiplier: %u, sys_div: %u, pre_div: %u, root_2x: %u, pclk_root_div: %u, pclk_manual: %u, pclk_div: %u", bypass, multiplier, sys_div, pre_div, root_2x, pclk_root_div, pclk_manual, pclk_div);

    calc_sysclk(sensor->xclk_freq_hz, bypass, multiplier, sys_div, pre_div, root_2x, pclk_root_div, pclk_manual, pclk_div);

@@ -432,14 +434,22 @@ static int set_framesize(sensor_t *sensor, framesize_t framesize)
    if (sensor->pixformat == PIXFORMAT_JPEG) {
        //10MHz PCLK
        uint8_t sys_mul = 200;
-        if(framesize < FRAMESIZE_QVGA){
+        if(framesize < FRAMESIZE_QVGA || sensor->xclk_freq_hz == 16000000){
            sys_mul = 160;
        } else if(framesize < FRAMESIZE_XGA){
            sys_mul = 180;
        }
        ret = set_pll(sensor, false, sys_mul, 4, 2, false, 2, true, 4);
+        //Set PLL: bypass: 0, multiplier: sys_mul, sys_div: 4, pre_div: 2, root_2x: 0, pclk_root_div: 2, pclk_manual: 1, pclk_div: 4
    } else {
-        ret = set_pll(sensor, false, 10, 1, 1, false, 1, true, 4);
+        //ret = set_pll(sensor, false, 8, 1, 1, false, 1, true, 4);
+        if (framesize > FRAMESIZE_HVGA) {
+            ret = set_pll(sensor, false, 10, 1, 2, false, 1, true, 2);
+        } else if (framesize >= FRAMESIZE_QVGA) {
+            ret = set_pll(sensor, false, 8, 1, 1, false, 1, true, 4);
+        } else {
+            ret = set_pll(sensor, false, 20, 1, 1, false, 1, true, 8);
+        }
    }

    if (ret == 0) {
@@ -1025,7 +1035,6 @@ static int _set_pll(sensor_t *sensor, int bypass, int multiplier, int sys_div, i
    return ret;
 }

-esp_err_t xclk_timer_conf(int ledc_timer, int xclk_freq_hz);
 static int set_xclk(sensor_t *sensor, int timer, int xclk)
 {
    int ret = 0;
@@ -1064,6 +1073,22 @@ static int init_status(sensor_t *sensor)
    return 0;
 }

+int ov5640_detect(int slv_addr, sensor_id_t *id)
+{
+    if (OV5640_SCCB_ADDR == slv_addr) {
+        uint8_t h = SCCB_Read16(slv_addr, 0x300A);
+        uint8_t l = SCCB_Read16(slv_addr, 0x300B);
+        uint16_t PID = (h<<8) | l;
+        if (OV5640_PID == PID) {
+            id->PID = PID;
+            return PID;
+        } else {
+            ESP_LOGI(TAG, "Mismatch PID=0x%x", PID);
+        }
+    }
+    return 0;
+}
+
 int ov5640_init(sensor_t *sensor)
 {
    sensor->reset = reset;
--- a/code/components/esp32-camera-master/sensors/ov7670.c
+++ b/code/components/esp32-camera-master/sensors/ov7670.c
@@ -45,7 +45,7 @@ static struct regval_list ov7670_default_regs[] = {
 	{CLKRC, 0x00},  
 	{DBLV,  0x4A},  

-    {COM10, COM10_VSYNC_NEG | COM10_PCLK_MASK},
+    {COM10, COM10_VSYNC_NEG | COM10_PCLK_FREE},

    /* Improve white balance */ 
 	{COM4, 0x40},  
@@ -393,6 +393,24 @@ static int init_status(sensor_t *sensor)
 static int set_dummy(sensor_t *sensor, int val){ return -1; }
 static int set_gainceiling_dummy(sensor_t *sensor, gainceiling_t val){ return -1; }

+int ov7670_detect(int slv_addr, sensor_id_t *id)
+{
+    if (OV7670_SCCB_ADDR == slv_addr) {
+        SCCB_Write(slv_addr, 0xFF, 0x01);//bank sensor
+        uint16_t PID = SCCB_Read(slv_addr, 0x0A);
+        if (OV7670_PID == PID) {
+            id->PID = PID;
+            id->VER = SCCB_Read(slv_addr, REG_VER);
+            id->MIDL = SCCB_Read(slv_addr, REG_MIDL);
+            id->MIDH = SCCB_Read(slv_addr, REG_MIDH);
+            return PID;
+        } else {
+            ESP_LOGI(TAG, "Mismatch PID=0x%x", PID);
+        }
+    }
+    return 0;
+}
+
 int ov7670_init(sensor_t *sensor)
 {
    // Set function pointers
--- a/code/components/esp32-camera-master/sensors/ov7725.c
+++ b/code/components/esp32-camera-master/sensors/ov7725.c
@@ -11,6 +11,7 @@
 #include <string.h>
 #include <stdio.h>
 #include "sccb.h"
+#include "xclk.h"
 #include "ov7725.h"
 #include "ov7725_regs.h"
 #include "freertos/FreeRTOS.h"
@@ -58,10 +59,10 @@ static const uint8_t default_regs[][2] = {
    {COM8,          0xF0},
    {COM6,          0xC5},
    {COM9,          0x11},
-    {COM10,         COM10_VSYNC_NEG | COM10_PCLK_MASK}, //Invert VSYNC and MASK PCLK
+    {COM10,         COM10_VSYNC_NEG | COM10_PCLK_FREE}, //Invert VSYNC and MASK PCLK
    {BDBASE,        0x7F},
    {DBSTEP,        0x03},
-    {AEW,           0x96},
+    {AEW,           0x75},
    {AEB,           0x64},
    {VPT,           0xA1},
    {EXHCL,         0x00},
@@ -493,7 +494,6 @@ static int set_gainceiling_dummy(sensor_t *sensor, gainceiling_t val){ return -1
 static int set_res_raw(sensor_t *sensor, int startX, int startY, int endX, int endY, int offsetX, int offsetY, int totalX, int totalY, int outputX, int outputY, bool scale, bool binning){return -1;}
 static int _set_pll(sensor_t *sensor, int bypass, int multiplier, int sys_div, int root_2x, int pre_div, int seld5, int pclk_manual, int pclk_div){return -1;}

-esp_err_t xclk_timer_conf(int ledc_timer, int xclk_freq_hz);
 static int set_xclk(sensor_t *sensor, int timer, int xclk)
 {
    int ret = 0;
@@ -502,6 +502,24 @@ static int set_xclk(sensor_t *sensor, int timer, int xclk)
    return ret;
 }

+int ov7725_detect(int slv_addr, sensor_id_t *id)
+{
+    if (OV7725_SCCB_ADDR == slv_addr) {
+        SCCB_Write(slv_addr, 0xFF, 0x01);//bank sensor
+        uint16_t PID = SCCB_Read(slv_addr, 0x0A);
+        if (OV7725_PID == PID) {
+            id->PID = PID;
+            id->VER = SCCB_Read(slv_addr, REG_VER);
+            id->MIDL = SCCB_Read(slv_addr, REG_MIDL);
+            id->MIDH = SCCB_Read(slv_addr, REG_MIDH);
+            return PID;
+        } else {
+            ESP_LOGI(TAG, "Mismatch PID=0x%x", PID);
+        }
+    }
+    return 0;
+}
+
 int ov7725_init(sensor_t *sensor)
 {
    // Set function pointers
--- a/code/components/esp32-camera-master/sensors/private_include/bf3005.h
+++ b/code/components/esp32-camera-master/sensors/private_include/bf3005.h
@@ -0,0 +1,33 @@
+/*
+ * This file is part of the OpenMV project.
+ * Copyright (c) 2013/2014 Ibrahim Abdelkader <i.abdalkader@gmail.com>
+ * This work is licensed under the MIT license, see the file LICENSE for details.
+ *
+ * BF3005 driver.
+ *
+ */
+#ifndef __BF3005_H__
+#define __BF3005_H__
+#include "sensor.h"
+
+/**
+ * @brief Detect sensor pid
+ *
+ * @param slv_addr SCCB address
+ * @param id Detection result
+ * @return
+ *     0:       Can't detect this sensor
+ *     Nonzero: This sensor has been detected
+ */
+int bf3005_detect(int slv_addr, sensor_id_t *id);
+
+/**
+ * @brief initialize sensor function pointers
+ *
+ * @param sensor pointer of sensor
+ * @return
+ *      Always 0
+ */
+int bf3005_init(sensor_t *sensor);
+
+#endif // __BF3005_H__
--- a/code/components/esp32-camera-master/sensors/private_include/bf3005_regs.h
+++ b/code/components/esp32-camera-master/sensors/private_include/bf3005_regs.h
@@ -0,0 +1,337 @@
+/*
+ * This file is part of the OpenMV project.
+ * Copyright (c) 2013/2014 Ibrahim Abdelkader <i.abdalkader@gmail.com>
+ * This work is licensed under the MIT license, see the file LICENSE for details.
+ *
+ * BF3005 register definitions.
+ */
+#ifndef __REG_REGS_H__
+#define __REG_REGS_H__
+#if 0
+#define GAIN                    0x00 /* AGC <20>C Gain control gain setting  */
+#define BLUE                    0x01 /* AWB <20>C Blue channel gain setting  */
+#define RED                     0x02 /* AWB <20>C Red channel gain setting   */
+#define GREEN                   0x03 /* AWB <20>C Green channel gain setting */
+#define BAVG                    0x05 /* U/B Average Level   */
+#define GAVG                    0x06 /* Y/Gb Average Level  */
+#define RAVG                    0x07 /* V/R Average Level   */
+#define AECH                    0x08 /* Exposure Value <20>C AEC MSBs */
+
+#define COM2                    0x09 /* Common Control 2 */
+#define COM2_SOFT_SLEEP         0x10 /* Soft sleep mode  */
+#define COM2_OUT_DRIVE_1x       0x00 /* Output drive capability 1x */
+#define COM2_OUT_DRIVE_2x       0x01 /* Output drive capability 2x */
+#define COM2_OUT_DRIVE_3x       0x02 /* Output drive capability 3x */
+#define COM2_OUT_DRIVE_4x       0x03 /* Output drive capability 4x */
+
+#define REG_PID                     0x0A /* Product ID Number MSB */
+#define REG_VER                     0x0B /* Product ID Number LSB */
+
+#define COM3                    0x0C /* Common Control 3                                        */
+#define COM3_VFLIP              0x80 /* Vertical flip image ON/OFF selection                    */
+#define COM3_MIRROR             0x40 /* Horizontal mirror image ON/OFF selection                */
+#define COM3_SWAP_BR            0x20 /* Swap B/R output sequence in RGB output mode             */
+#define COM3_SWAP_YUV           0x10 /* Swap Y/UV output sequence in YUV output mode            */
+#define COM3_SWAP_MSB           0x08 /* Swap output MSB/LSB                                     */
+#define COM3_TRI_CLOCK          0x04 /* Tri-state option for output clock at power-down period  */
+#define COM3_TRI_DATA           0x02 /* Tri-state option for output data at power-down period   */
+#define COM3_COLOR_BAR          0x01 /* Sensor color bar test pattern output enable             */
+#define COM3_SET_CBAR(r, x)     ((r&0xFE)|((x&1)<<0))
+#define COM3_SET_MIRROR(r, x)   ((r&0xBF)|((x&1)<<6))
+#define COM3_SET_FLIP(r, x)     ((r&0x7F)|((x&1)<<7))
+
+#define COM4                    0x0D /* Common Control 4         */
+#define COM4_PLL_BYPASS         0x00 /* Bypass PLL               */
+#define COM4_PLL_4x             0x40 /* PLL frequency 4x         */
+#define COM4_PLL_6x             0x80 /* PLL frequency 6x         */
+#define COM4_PLL_8x             0xc0 /* PLL frequency 8x         */
+#define COM4_AEC_FULL           0x00 /* AEC evaluate full window */
+#define COM4_AEC_1_2            0x10 /* AEC evaluate 1/2 window  */
+#define COM4_AEC_1_4            0x20 /* AEC evaluate 1/4 window  */
+#define COM4_AEC_2_3            0x30 /* AEC evaluate 2/3 window  */
+
+#define COM5                    0x0E /* Common Control 5 */
+#define COM5_AFR                0x80 /* Auto frame rate control ON/OFF selection (night mode) */
+#define COM5_AFR_SPEED          0x40 /* Auto frame rate control speed selection */
+#define COM5_AFR_0              0x00 /* No reduction of frame rate          */
+#define COM5_AFR_1_2            0x10 /* Max reduction to 1/2 frame rate     */
+#define COM5_AFR_1_4            0x20 /* Max reduction to 1/4 frame rate     */
+#define COM5_AFR_1_8            0x30 /* Max reduction to 1/8 frame rate     */
+#define COM5_AFR_4x             0x04 /* Add frame when AGC reaches 4x gain  */
+#define COM5_AFR_8x             0x08 /* Add frame when AGC reaches 8x gain  */
+#define COM5_AFR_16x            0x0c /* Add frame when AGC reaches 16x gain */
+#define COM5_AEC_NO_LIMIT       0x01 /* No limit to AEC increase step       */
+
+#define COM6                    0x0F /* Common Control 6 */
+#define COM6_AUTO_WINDOW        0x01 /* Auto window setting ON/OFF selection when format changes */
+
+#define AEC                     0x10 /* AEC[7:0] (see register AECH for AEC[15:8]) */
+#define CLKRC                   0x11 /* Internal Clock */
+
+#define COM7                    0x12 /* Common Control 7         */
+#define COM7_RESET              0x80 /* SCCB Register Reset      */
+#define COM7_RES_VGA            0x00 /* Resolution VGA           */
+#define COM7_RES_QVGA           0x40 /* Resolution QVGA          */
+#define COM7_BT656              0x20 /* BT.656 protocol ON/OFF   */
+#define COM7_SENSOR_RAW         0x10 /* Sensor RAW               */
+#define COM7_FMT_GBR422         0x00 /* RGB output format GBR422 */
+#define COM7_FMT_RGB565         0x04 /* RGB output format RGB565 */
+#define COM7_FMT_RGB555         0x08 /* RGB output format RGB555 */
+#define COM7_FMT_RGB444         0x0C /* RGB output format RGB444 */
+#define COM7_FMT_YUV            0x00 /* Output format YUV        */
+#define COM7_FMT_P_BAYER        0x01 /* Output format Processed Bayer RAW */
+#define COM7_FMT_RGB            0x02 /* Output format RGB        */
+#define COM7_FMT_R_BAYER        0x03 /* Output format Bayer RAW  */
+#define COM7_SET_FMT(r, x)      ((r&0xFC)|((x&0x3)<<0))
+#define COM7_SET_RGB(r, x)      ((r&0xF0)|(x&0x0C)|COM7_FMT_RGB)
+
+#define COM8                    0x13 /* Common Control 8                */
+#define COM8_FAST_AUTO          0x80 /* Enable fast AGC/AEC algorithm   */
+#define COM8_STEP_VSYNC         0x00 /* AEC - Step size limited to vertical blank */
+#define COM8_STEP_UNLIMIT       0x40 /* AEC - Step size unlimited step size       */
+#define COM8_BANDF_EN           0x20 /* Banding filter ON/OFF */
+#define COM8_AEC_BANDF          0x10 /* Enable AEC below banding value */
+#define COM8_AEC_FINE_EN        0x08 /* Fine AEC ON/OFF control */
+#define COM8_AGC_EN             0x04 /* AGC Enable */
+#define COM8_AWB_EN             0x02 /* AWB Enable */
+#define COM8_AEC_EN             0x01 /* AEC Enable */
+#define COM8_SET_AGC(r, x)      ((r&0xFB)|((x&0x1)<<2))
+#define COM8_SET_AWB(r, x)      ((r&0xFD)|((x&0x1)<<1))
+#define COM8_SET_AEC(r, x)      ((r&0xFE)|((x&0x1)<<0))
+
+#define COM9                    0x14 /* Common Control 9 */
+#define COM9_HISTO_AVG          0x80 /* Histogram or average based AEC/AGC selection */
+#define COM9_AGC_GAIN_2x        0x00 /* Automatic Gain Ceiling 2x  */
+#define COM9_AGC_GAIN_4x        0x10 /* Automatic Gain Ceiling 4x  */
+#define COM9_AGC_GAIN_8x        0x20 /* Automatic Gain Ceiling 8x  */
+#define COM9_AGC_GAIN_16x       0x30 /* Automatic Gain Ceiling 16x */
+#define COM9_AGC_GAIN_32x       0x40 /* Automatic Gain Ceiling 32x */
+#define COM9_DROP_VSYNC         0x04 /* Drop VSYNC output of corrupt frame */
+#define COM9_DROP_HREF          0x02 /* Drop HREF output of corrupt frame  */
+#define COM9_SET_AGC(r, x)      ((r&0x8F)|((x&0x07)<<4))
+
+#define COM10                   0x15 /* Common Control 10 */
+#define COM10_NEGATIVE          0x80 /* Output negative data */
+#define COM10_HSYNC_EN          0x40 /* HREF changes to HSYNC */
+#define COM10_PCLK_FREE         0x00 /* PCLK output option: free running PCLK */
+#define COM10_PCLK_MASK         0x20 /* PCLK output option: masked during horizontal blank  */
+#define COM10_PCLK_REV          0x10 /* PCLK reverse */
+#define COM10_HREF_REV          0x08 /* HREF reverse */
+#define COM10_VSYNC_FALLING     0x00 /* VSYNC changes on falling edge of PCLK */
+#define COM10_VSYNC_RISING      0x04 /* VSYNC changes on rising edge of PCLK */
+#define COM10_VSYNC_NEG         0x02 /* VSYNC negative */
+#define COM10_OUT_RANGE_8       0x01 /* Output data range: Full range */
+#define COM10_OUT_RANGE_10      0x00 /* Output data range: Data from [10] to [F0] (8 MSBs) */
+
+#define REG16                   0x16 /* Register 16 */
+#define REG16_BIT_SHIFT         0x80 /* Bit shift test pattern options */
+#define HSTART                  0x17 /* Horizontal Frame (HREF column) Start 8 MSBs (2 LSBs are at HREF[5:4]) */
+#define HSIZE                   0x18 /* Horizontal Sensor Size (2 LSBs are at HREF[1:0]) */
+#define VSTART                  0x19 /* Vertical Frame (row) Start 8 MSBs (1 LSB is at HREF[6]) */
+#define VSIZE                   0x1A /* Vertical Sensor Size (1 LSB is at HREF[2]) */
+#define PSHFT                   0x1B /* Data Format - Pixel Delay Select */
+#define REG_MIDH                    0x1C /* Manufacturer ID Byte <20>C High */
+#define REG_MIDL                    0x1D /* Manufacturer ID Byte <20>C Low */
+#define LAEC                    0x1F /* Fine AEC Value - defines exposure value less than one row period */
+
+#define COM11                   0x20 /* Common Control 11 */
+#define COM11_SNGL_FRAME_EN     0x02 /* Single frame ON/OFF selection */
+#define COM11_SNGL_XFR_TRIG     0x01 /* Single frame transfer trigger */
+
+#define BDBASE                  0x22 /* Banding Filter Minimum AEC Value */
+#define DBSTEP                  0x23 /* Banding Filter Maximum Step */
+#define AEW                     0x24 /* AGC/AEC - Stable Operating Region (Upper Limit) */
+#define AEB                     0x25 /* AGC/AEC - Stable Operating Region (Lower Limit) */
+#define VPT                     0x26 /* AGC/AEC Fast Mode Operating Region */
+#define REG28                   0x28 /* Selection on the number of dummy rows, N */
+#define HOUTSIZE                0x29 /* Horizontal Data Output Size MSBs (2 LSBs at register EXHCH[1:0]) */
+#define EXHCH                   0x2A /* Dummy Pixel Insert MSB */
+#define EXHCL                   0x2B /* Dummy Pixel Insert LSB */
+#define VOUTSIZE                0x2C /* Vertical Data Output Size MSBs (LSB at register EXHCH[2])       */
+#define ADVFL                   0x2D /* LSB of Insert Dummy Rows in Vertical Sync (1 bit equals 1 row)  */
+#define ADVFH                   0x2E /* MSB of Insert Dummy Rows in Vertical Sync */
+#define YAVE                    0x2F /* Y/G Channel Average Value */
+#define LUMHTH                  0x30 /* Histogram AEC/AGC Luminance High Level Threshold */
+#define LUMLTH                  0x31 /* Histogram AEC/AGC Luminance Low Level Threshold  */
+#define HREF                    0x32 /* Image Start and Size Control */
+#define DM_LNL                  0x33 /* Dummy Row Low 8 Bits  */
+#define DM_LNH                  0x34 /* Dummy Row High 8 Bits */
+#define ADOFF_B                 0x35 /* AD Offset Compensation Value for B Channel  */
+#define ADOFF_R                 0x36 /* AD Offset Compensation Value for R Channel  */
+#define ADOFF_GB                0x37 /* AD Offset Compensation Value for GB Channel */
+#define ADOFF_GR                0x38 /* AD Offset Compensation Value for GR Channel */
+#define OFF_B                   0x39 /* AD Offset Compensation Value for B Channel  */
+#define OFF_R                   0x3A /* AD Offset Compensation Value for R Channel  */
+#define OFF_GB                  0x3B /* AD Offset Compensation Value for GB Channel */
+#define OFF_GR                  0x3C /* AD Offset Compensation Value for GR Channel */
+#define COM12                   0x3D /* DC offset compensation for analog process */
+
+#define COM13                   0x3E /* Common Control 13 */
+#define COM13_BLC_EN            0x80 /* BLC enable */
+#define COM13_ADC_EN            0x40 /* ADC channel BLC ON/OFF control */
+#define COM13_ANALOG_BLC        0x20 /* Analog processing channel BLC ON/OFF control */
+#define COM13_ABLC_GAIN_EN      0x04 /* ABLC gain trigger enable */
+
+#define COM14                   0x3F /* Common Control 14 */
+#define COM15                   0x40 /* Common Control 15 */
+#define COM16                   0x41 /* Common Control 16 */
+#define TGT_B                   0x42 /* BLC Blue Channel Target Value   */
+#define TGT_R                   0x43 /* BLC Red Channel Target Value    */
+#define TGT_GB                  0x44 /* BLC Gb Channel Target Value     */
+#define TGT_GR                  0x45 /* BLC Gr Channel Target Value     */
+
+#define LC_CTR                  0x46 /* Lens Correction Control */
+#define LC_CTR_RGB_COMP_1       0x00 /* R, G, and B channel compensation coefficient is set by LC_COEF (0x49) */
+#define LC_CTR_RGB_COMP_3       0x04 /* R, G, and B channel compensation coefficient is set by registers
+                                        LC_COEFB (0x4B), LC_COEF (0x49), and LC_COEFR (0x4C), respectively */
+#define LC_CTR_EN               0x01 /* Lens correction enable */
+#define LC_XC                   0x47 /* X Coordinate of Lens Correction Center Relative to Array Center */
+#define LC_YC                   0x48 /* Y Coordinate of Lens Correction Center Relative to Array Center */
+#define LC_COEF                 0x49 /* Lens Correction Coefficient */
+#define LC_RADI                 0x4A /* Lens Correction Radius */
+#define LC_COEFB                0x4B /* Lens Correction B Channel Compensation Coefficient */
+#define LC_COEFR                0x4C /* Lens Correction R Channel Compensation Coefficient */
+
+#define FIXGAIN                 0x4D /* Analog Fix Gain Amplifier */
+#define AREF0                   0x4E /* Sensor Reference Control */
+#define AREF1                   0x4F /* Sensor Reference Current Control */
+#define AREF2                   0x50 /* Analog Reference Control */
+#define AREF3                   0x51 /* ADC Reference Control */
+#define AREF4                   0x52 /* ADC Reference Control */
+#define AREF5                   0x53 /* ADC Reference Control */
+#define AREF6                   0x54 /* Analog Reference Control */
+#define AREF7                   0x55 /* Analog Reference Control */
+#define UFIX                    0x60 /* U Channel Fixed Value Output */
+#define VFIX                    0x61 /* V Channel Fixed Value Output */
+#define AWBB_BLK                0x62 /* AWB Option for Advanced AWB  */
+
+#define AWB_CTRL0               0x63 /* AWB Control Byte 0   */
+#define AWB_CTRL0_GAIN_EN       0x80 /* AWB gain enable      */
+#define AWB_CTRL0_CALC_EN       0x40 /* AWB calculate enable */
+#define AWB_CTRL0_WBC_MASK      0x0F /* WBC threshold 2      */
+
+#define DSP_CTRL1               0x64 /* DSP Control Byte 1                  */
+#define DSP_CTRL1_FIFO_EN       0x80 /* FIFO enable/disable selection       */
+#define DSP_CTRL1_UV_EN         0x40 /* UV adjust function ON/OFF selection */
+#define DSP_CTRL1_SDE_EN        0x20 /* SDE enable                          */
+#define DSP_CTRL1_MTRX_EN       0x10 /* Color matrix ON/OFF selection       */
+#define DSP_CTRL1_INTRP_EN      0x08 /* Interpolation ON/OFF selection      */
+#define DSP_CTRL1_GAMMA_EN      0x04 /* Gamma function ON/OFF selection     */
+#define DSP_CTRL1_BLACK_EN      0x02 /* Black defect auto correction ON/OFF */
+#define DSP_CTRL1_WHITE_EN      0x01 /* White defect auto correction ON/OFF */
+
+#define DSP_CTRL2               0x65 /* DSP Control Byte 2          */
+#define DSP_CTRL2_VDCW_EN       0x08 /* Vertical DCW enable         */
+#define DSP_CTRL2_HDCW_EN       0x04 /* Horizontal DCW enable       */
+#define DSP_CTRL2_VZOOM_EN      0x02 /* Vertical zoom out enable    */
+#define DSP_CTRL2_HZOOM_EN      0x01 /* Horizontal zoom out enable  */
+
+#define DSP_CTRL3               0x66 /* DSP Control Byte 3                      */
+#define DSP_CTRL3_UV_EN         0x80 /* UV output sequence option               */
+#define DSP_CTRL3_CBAR_EN       0x20 /* DSP color bar ON/OFF selection          */
+#define DSP_CTRL3_FIFO_EN       0x08 /* FIFO power down ON/OFF selection        */
+#define DSP_CTRL3_SCAL1_PWDN    0x04 /* Scaling module power down control 1     */
+#define DSP_CTRL3_SCAL2_PWDN    0x02 /* Scaling module power down control 2     */
+#define DSP_CTRL3_INTRP_PWDN    0x01 /* Interpolation module power down control */
+#define DSP_CTRL3_SET_CBAR(r, x)    ((r&0xDF)|((x&1)<<5))
+
+
+#define DSP_CTRL4               0x67 /* DSP Control Byte 4          */
+#define DSP_CTRL4_YUV_RGB       0x00 /* Output selection YUV or RGB */
+#define DSP_CTRL4_RAW8          0x02 /* Output selection RAW8       */
+#define DSP_CTRL4_RAW10         0x03 /* Output selection RAW10      */
+
+
+#define AWB_BIAS                0x68 /* AWB BLC Level Clip */
+#define AWB_CTRL1               0x69 /* AWB Control 1 */
+#define AWB_CTRL2               0x6A /* AWB Control 2 */
+
+#define AWB_CTRL3               0x6B /* AWB Control 3 */
+#define AWB_CTRL3_ADVANCED      0x80 /* AWB mode select - Advanced AWB */
+#define AWB_CTRL3_SIMPLE        0x00 /* AWB mode select - Simple AWB */
+
+#define AWB_CTRL4               0x6C /* AWB Control 4  */
+#define AWB_CTRL5               0x6D /* AWB Control 5  */
+#define AWB_CTRL6               0x6E /* AWB Control 6  */
+#define AWB_CTRL7               0x6F /* AWB Control 7  */
+#define AWB_CTRL8               0x70 /* AWB Control 8  */
+#define AWB_CTRL9               0x71 /* AWB Control 9  */
+#define AWB_CTRL10              0x72 /* AWB Control 10 */
+#define AWB_CTRL11              0x73 /* AWB Control 11 */
+#define AWB_CTRL12              0x74 /* AWB Control 12 */
+#define AWB_CTRL13              0x75 /* AWB Control 13 */
+#define AWB_CTRL14              0x76 /* AWB Control 14 */
+#define AWB_CTRL15              0x77 /* AWB Control 15 */
+#define AWB_CTRL16              0x78 /* AWB Control 16 */
+#define AWB_CTRL17              0x79 /* AWB Control 17 */
+#define AWB_CTRL18              0x7A /* AWB Control 18 */
+#define AWB_CTRL19              0x7B /* AWB Control 19 */
+#define AWB_CTRL20              0x7C /* AWB Control 20 */
+#define AWB_CTRL21              0x7D /* AWB Control 21 */
+#define GAM1                    0x7E /* Gamma Curve 1st Segment Input End Point 0x04 Output Value */
+#define GAM2                    0x7F /* Gamma Curve 2nd Segment Input End Point 0x08 Output Value */
+#define GAM3                    0x80 /* Gamma Curve 3rd Segment Input End Point 0x10 Output Value */
+#define GAM4                    0x81 /* Gamma Curve 4th Segment Input End Point 0x20 Output Value */
+#define GAM5                    0x82 /* Gamma Curve 5th Segment Input End Point 0x28 Output Value */
+#define GAM6                    0x83 /* Gamma Curve 6th Segment Input End Point 0x30 Output Value */
+#define GAM7                    0x84 /* Gamma Curve 7th Segment Input End Point 0x38 Output Value */
+#define GAM8                    0x85 /* Gamma Curve 8th Segment Input End Point 0x40 Output Value */
+#define GAM9                    0x86 /* Gamma Curve 9th Segment Input End Point 0x48 Output Value */
+#define GAM10                   0x87 /* Gamma Curve 10th Segment Input End Point 0x50 Output Value */
+#define GAM11                   0x88 /* Gamma Curve 11th Segment Input End Point 0x60 Output Value */
+#define GAM12                   0x89 /* Gamma Curve 12th Segment Input End Point 0x70 Output Value */
+#define GAM13                   0x8A /* Gamma Curve 13th Segment Input End Point 0x90 Output Value */
+#define GAM14                   0x8B /* Gamma Curve 14th Segment Input End Point 0xB0 Output Value */
+#define GAM15                   0x8C /* Gamma Curve 15th Segment Input End Point 0xD0 Output Value */
+#define SLOP                    0x8D /* Gamma Curve Highest Segment Slope */
+#define DNSTH                   0x8E /* De-noise Threshold */
+#define EDGE0                   0x8F /* Edge Enhancement Strength Control */
+#define EDGE1                   0x90 /* Edge Enhancement Threshold Control */
+#define DNSOFF                  0x91 /* Auto De-noise Threshold Control */
+#define EDGE2                   0x92 /* Edge Enhancement Strength Upper Limit */
+#define EDGE3                   0x93 /* Edge Enhancement Strength Upper Limit */
+#define MTX1                    0x94 /* Matrix Coefficient 1 */
+#define MTX2                    0x95 /* Matrix Coefficient 2 */
+#define MTX3                    0x96 /* Matrix Coefficient 3 */
+#define MTX4                    0x97 /* Matrix Coefficient 4 */
+#define MTX5                    0x98 /* Matrix Coefficient 5 */
+#define MTX6                    0x99 /* Matrix Coefficient 6 */
+
+#define MTX_CTRL                0x9A /* Matrix Control */
+#define MTX_CTRL_DBL_EN         0x80 /* Matrix double ON/OFF selection */
+
+#define BRIGHTNESS              0x9B /* Brightness Control */
+#define CONTRAST                0x9C /* Contrast Gain */
+#define UVADJ0                  0x9E /* Auto UV Adjust Control 0 */
+#define UVADJ1                  0x9F /* Auto UV Adjust Control 1 */
+#define SCAL0                   0xA0 /* DCW Ratio Control */
+#define SCAL1                   0xA1 /* Horizontal Zoom Out Control */
+#define SCAL2                   0xA2 /* Vertical Zoom Out Control */
+#define FIFODLYM                0xA3 /* FIFO Manual Mode Delay Control */
+#define FIFODLYA                0xA4 /* FIFO Auto Mode Delay Control */
+
+#define SDE                     0xA6 /* Special Digital Effect Control  */
+#define SDE_NEGATIVE_EN         0x40 /* Negative image enable           */
+#define SDE_GRAYSCALE_EN        0x20 /* Gray scale image enable         */
+#define SDE_V_FIXED_EN          0x10 /* V fixed value enable            */
+#define SDE_U_FIXED_EN          0x08 /* U fixed value enable            */
+#define SDE_CONT_BRIGHT_EN      0x04 /* Contrast/Brightness enable      */
+#define SDE_SATURATION_EN       0x02 /* Saturation enable               */
+#define SDE_HUE_EN              0x01 /* Hue enable                      */
+
+#define USAT                    0xA7 /* U Component Saturation Gain     */
+#define VSAT                    0xA8 /* V Component Saturation Gain     */
+#define HUECOS                  0xA9 /* Cosine value <20><> 0x80             */
+#define HUESIN                  0xAA /* Sine value <20><> 0x80               */
+#define SIGN_BIT                0xAB /* Sign Bit for Hue and Brightness */
+
+#define DSPAUTO                 0xAC /* DSP Auto Function ON/OFF Control */
+#define DSPAUTO_AWB_EN          0x80 /* AWB auto threshold control */
+#define DSPAUTO_DENOISE_EN      0x40 /* De-noise auto threshold control */
+#define DSPAUTO_EDGE_EN         0x20 /* Sharpness (edge enhancement) auto strength control */
+#define DSPAUTO_UV_EN           0x10 /* UV adjust auto slope control */
+#define DSPAUTO_SCAL0_EN        0x08 /* Auto scaling factor control (register SCAL0 (0xA0)) */
+#define DSPAUTO_SCAL1_EN        0x04 /* Auto scaling factor control (registers SCAL1 (0xA1 and SCAL2 (0xA2))*/
+#define SET_REG(reg, x)         (##reg_DEFAULT|x)
+#endif //__REG_REGS_H__
+#endif
--- a/code/components/esp32-camera-master/sensors/private_include/gc0308.h
+++ b/code/components/esp32-camera-master/sensors/private_include/gc0308.h
@@ -0,0 +1,31 @@
+#pragma once
+
+#include "sensor.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+/**
+ * @brief Detect sensor pid
+ *
+ * @param slv_addr SCCB address
+ * @param id Detection result
+ * @return
+ *     0:       Can't detect this sensor
+ *     Nonzero: This sensor has been detected
+ */
+int gc0308_detect(int slv_addr, sensor_id_t *id);
+
+/**
+ * @brief initialize sensor function pointers
+ *
+ * @param sensor pointer of sensor
+ * @return
+ *      Always 0
+ */
+int gc0308_init(sensor_t *sensor);
+
+#ifdef __cplusplus
+}
+#endif
--- a/code/components/esp32-camera-master/sensors/private_include/gc0308_regs.h
+++ b/code/components/esp32-camera-master/sensors/private_include/gc0308_regs.h
@@ -0,0 +1,25 @@
+/*
+ * GC0308 register definitions.
+ */
+#ifndef __GC0308_REG_REGS_H__
+#define __GC0308_REG_REGS_H__
+
+#define RESET_RELATED   0xfe    // Bit[7]: Software reset 
+                                // Bit[6:5]: NA
+                                // Bit[4]: CISCTL_restart_n
+                                // Bit[3:1]: NA
+                                // Bit[0]: page select
+                                //  0:page0
+                                //  1:page1
+
+
+// page0:
+
+
+
+/**
+ * @brief register value
+ */
+
+
+#endif // __GC0308_REG_REGS_H__
--- a/code/components/esp32-camera-master/sensors/private_include/gc0308_settings.h
+++ b/code/components/esp32-camera-master/sensors/private_include/gc0308_settings.h
@@ -0,0 +1,245 @@
+#ifndef _GC0308_SETTINGS_H_
+#define _GC0308_SETTINGS_H_
+
+#include <stdint.h>
+
+#define REG_DLY 0xffff
+#define REGLIST_TAIL 0x0000 /* Array end token */
+
+static const uint16_t gc0308_sensor_default_regs[][2] = {
+    {0xfe, 0x00},
+    {0xec, 0x20},
+    {0x05, 0x00},
+    {0x06, 0x00},
+    {0x07, 0x00},
+    {0x08, 0x00},
+    {0x09, 0x01},
+    {0x0a, 0xe8},
+    {0x0b, 0x02},
+    {0x0c, 0x88},
+    {0x0d, 0x02},
+    {0x0e, 0x02},
+    {0x10, 0x26},
+    {0x11, 0x0d},
+    {0x12, 0x2a},
+    {0x13, 0x00},
+    {0x14, 0x11},
+    {0x15, 0x0a},
+    {0x16, 0x05},
+    {0x17, 0x01},
+    {0x18, 0x44},
+    {0x19, 0x44},
+    {0x1a, 0x2a},
+    {0x1b, 0x00},
+    {0x1c, 0x49},
+    {0x1d, 0x9a},
+    {0x1e, 0x61},
+    {0x1f, 0x00}, //pad drv <=24MHz, use 0x00 is ok
+    {0x20, 0x7f},
+    {0x21, 0xfa},
+    {0x22, 0x57},
+    {0x24, 0xa2},   //YCbYCr
+    {0x25, 0x0f},
+    {0x26, 0x03}, // 0x01
+    {0x28, 0x00},
+    {0x2d, 0x0a},
+    {0x2f, 0x01},
+    {0x30, 0xf7},
+    {0x31, 0x50},
+    {0x32, 0x00},
+    {0x33, 0x28},
+    {0x34, 0x2a},
+    {0x35, 0x28},
+    {0x39, 0x04},
+    {0x3a, 0x20},
+    {0x3b, 0x20},
+    {0x3c, 0x00},
+    {0x3d, 0x00},
+    {0x3e, 0x00},
+    {0x3f, 0x00},
+    {0x50, 0x14}, // 0x14
+    {0x52, 0x41},
+    {0x53, 0x80},
+    {0x54, 0x80},
+    {0x55, 0x80},
+    {0x56, 0x80},
+    {0x8b, 0x20},
+    {0x8c, 0x20},
+    {0x8d, 0x20},
+    {0x8e, 0x14},
+    {0x8f, 0x10},
+    {0x90, 0x14},
+    {0x91, 0x3c},
+    {0x92, 0x50},
+//{0x8b,0x10},
+//{0x8c,0x10},
+//{0x8d,0x10},
+//{0x8e,0x10},
+//{0x8f,0x10},
+//{0x90,0x10},
+//{0x91,0x3c},
+//{0x92,0x50},
+    {0x5d, 0x12},
+    {0x5e, 0x1a},
+    {0x5f, 0x24},
+    {0x60, 0x07},
+    {0x61, 0x15},
+    {0x62, 0x08}, // 0x08
+    {0x64, 0x03}, // 0x03
+    {0x66, 0xe8},
+    {0x67, 0x86},
+    {0x68, 0x82},
+    {0x69, 0x18},
+    {0x6a, 0x0f},
+    {0x6b, 0x00},
+    {0x6c, 0x5f},
+    {0x6d, 0x8f},
+    {0x6e, 0x55},
+    {0x6f, 0x38},
+    {0x70, 0x15},
+    {0x71, 0x33},
+    {0x72, 0xdc},
+    {0x73, 0x00},
+    {0x74, 0x02},
+    {0x75, 0x3f},
+    {0x76, 0x02},
+    {0x77, 0x38}, // 0x47
+    {0x78, 0x88},
+    {0x79, 0x81},
+    {0x7a, 0x81},
+    {0x7b, 0x22},
+    {0x7c, 0xff},
+    {0x93, 0x48}, //color matrix default
+    {0x94, 0x02},
+    {0x95, 0x07},
+    {0x96, 0xe0},
+    {0x97, 0x40},
+    {0x98, 0xf0},
+    {0xb1, 0x40},
+    {0xb2, 0x40},
+    {0xb3, 0x40}, //0x40
+    {0xb6, 0xe0},
+    {0xbd, 0x38},
+    {0xbe, 0x36},
+    {0xd0, 0xCB},
+    {0xd1, 0x10},
+    {0xd2, 0x90},
+    {0xd3, 0x48},
+    {0xd5, 0xF2},
+    {0xd6, 0x16},
+    {0xdb, 0x92},
+    {0xdc, 0xA5},
+    {0xdf, 0x23},
+    {0xd9, 0x00},
+    {0xda, 0x00},
+    {0xe0, 0x09},
+    {0xed, 0x04},
+    {0xee, 0xa0},
+    {0xef, 0x40},
+    {0x80, 0x03},
+
+    {0x9F, 0x10},
+    {0xA0, 0x20},
+    {0xA1, 0x38},
+    {0xA2, 0x4e},
+    {0xA3, 0x63},
+    {0xA4, 0x76},
+    {0xA5, 0x87},
+    {0xA6, 0xa2},
+    {0xA7, 0xb8},
+    {0xA8, 0xca},
+    {0xA9, 0xd8},
+    {0xAA, 0xe3},
+    {0xAB, 0xeb},
+    {0xAC, 0xf0},
+    {0xAD, 0xF8},
+    {0xAE, 0xFd},
+    {0xAF, 0xFF},
+
+    {0xc0, 0x00},
+    {0xc1, 0x10},
+    {0xc2, 0x1c},
+    {0xc3, 0x30},
+    {0xc4, 0x43},
+    {0xc5, 0x54},
+    {0xc6, 0x65},
+    {0xc7, 0x75},
+    {0xc8, 0x93},
+    {0xc9, 0xB0},
+    {0xca, 0xCB},
+    {0xcb, 0xE6},
+    {0xcc, 0xFF},
+    {0xf0, 0x02},
+    {0xf1, 0x01},
+    {0xf2, 0x02},
+    {0xf3, 0x30},
+    {0xf7, 0x04},
+    {0xf8, 0x02},
+    {0xf9, 0x9f},
+    {0xfa, 0x78},
+    {0xfe, 0x01},
+    {0x00, 0xf5},
+    {0x02, 0x20},
+    {0x04, 0x10},
+    {0x05, 0x08},
+    {0x06, 0x20},
+    {0x08, 0x0a},
+    {0x0a, 0xa0},
+    {0x0b, 0x60},
+    {0x0c, 0x08},
+    {0x0e, 0x44},
+    {0x0f, 0x32},
+    {0x10, 0x41},
+    {0x11, 0x37},
+    {0x12, 0x22},
+    {0x13, 0x19},
+    {0x14, 0x44},
+    {0x15, 0x44},
+    {0x16, 0xc2},
+    {0x17, 0xA8},
+    {0x18, 0x18},
+    {0x19, 0x50},
+    {0x1a, 0xd8},
+    {0x1b, 0xf5},
+    {0x70, 0x40},
+    {0x71, 0x58},
+    {0x72, 0x30},
+    {0x73, 0x48},
+    {0x74, 0x20},
+    {0x75, 0x60},
+    {0x77, 0x20},
+    {0x78, 0x32},
+    {0x30, 0x03},
+    {0x31, 0x40},
+    {0x32, 0x10},
+    {0x33, 0xe0},
+    {0x34, 0xe0},
+    {0x35, 0x00},
+    {0x36, 0x80},
+    {0x37, 0x00},
+    {0x38, 0x04},
+    {0x39, 0x09},
+    {0x3a, 0x12},
+    {0x3b, 0x1C},
+    {0x3c, 0x28},
+    {0x3d, 0x31},
+    {0x3e, 0x44},
+    {0x3f, 0x57},
+    {0x40, 0x6C},
+    {0x41, 0x81},
+    {0x42, 0x94},
+    {0x43, 0xA7},
+    {0x44, 0xB8},
+    {0x45, 0xD6},
+    {0x46, 0xEE},
+    {0x47, 0x0d},
+    {0x62, 0xf7},
+    {0x63, 0x68},
+    {0x64, 0xd3},
+    {0x65, 0xd3},
+    {0x66, 0x60},
+    {0xfe, 0x00},
+    {REGLIST_TAIL, 0x00},
+};
+
+#endif
--- a/code/components/esp32-camera-master/sensors/private_include/gc032a.h
+++ b/code/components/esp32-camera-master/sensors/private_include/gc032a.h
@@ -0,0 +1,31 @@
+/*
+ *
+ * GC032A driver.
+ *
+ */
+#ifndef __GC032A_H__
+#define __GC032A_H__
+
+#include "sensor.h"
+
+/**
+ * @brief Detect sensor pid
+ *
+ * @param slv_addr SCCB address
+ * @param id Detection result
+ * @return
+ *     0:       Can't detect this sensor
+ *     Nonzero: This sensor has been detected
+ */
+int gc032a_detect(int slv_addr, sensor_id_t *id);
+
+/**
+ * @brief initialize sensor function pointers
+ *
+ * @param sensor pointer of sensor
+ * @return
+ *      Always 0
+ */
+int gc032a_init(sensor_t *sensor);
+
+#endif // __GC032A_H__
--- a/code/components/esp32-camera-master/sensors/private_include/gc032a_regs.h
+++ b/code/components/esp32-camera-master/sensors/private_include/gc032a_regs.h
@@ -0,0 +1,82 @@
+/*
+ * GC032A register definitions.
+ */
+#ifndef __GC032A_REG_REGS_H__
+#define __GC032A_REG_REGS_H__
+
+#define SENSOR_ID_HIGH 0XF0
+#define SENSOR_ID_LOW 0XF1
+#define PAD_VB_HIZ_MODE 0XF2
+#define SYNC_OUTPUT 0XF3
+#define I2C_CONFIG 0XF4
+#define PLL_MODE1 0XF7
+#define PLL_MODE2 0XF8
+#define CM_MODE 0XF9
+#define ISP_DIV_MODE 0XFA
+#define I2C_DEVICE_ID 0XFB
+#define ANALOG_PWC 0XFC
+#define ISP_DIV_MODE2 0XFD
+#define RESET_RELATED   0XFE    // Bit[7]: Software reset 
+                                // Bit[6]: cm reset 
+                                // Bit[5]: spi reset 
+                                // Bit[4]: CISCTL_restart_n
+                                // Bit[3]: PLL_rst
+                                // Bit[2:0]: page select
+                                //  000:page0
+                                //  001:page1
+                                //  010:page2
+                                //  011:page3
+
+//----page0-----------------------------
+#define P0_EXPOSURE_HIGH 0X03
+#define P0_EXPOSURE_LOW 0X04
+#define P0_HB_HIGH 0X05
+#define P0_HB_LOW 0X06
+#define P0_VB_HIGH 0X07
+#define P0_VB_LOW 0X08
+#define P0_ROW_START_HIGH 0X09
+#define P0_ROW_START_LOW 0X0A
+#define P0_COLUMN_START_HIGH 0X0B
+#define P0_COLUMN_START_LOW 0X0C
+#define P0_WINDOW_HEIGHT_HIGH 0X0D
+#define P0_WINDOW_HEIGHT_LOW 0X0E
+#define P0_WINDOW_WIDTH_HIGH 0X0F
+#define P0_WINDOW_WIDTH_LOW 0X10
+#define P0_SH_DELAY 0X11
+#define P0_VS_ST 0X12
+#define P0_VS_ET 0X13
+#define P0_CISCTL_MODE1 0X17
+
+#define P0_BLOCK_ENABLE_1 0X40
+#define P0_AAAA_ENABLE 0X42
+#define P0_SPECIAL_EFFECT 0X43
+#define P0_SYNC_MODE 0X46
+#define P0_GAIN_CODE 0X48
+#define P0_DEBUG_MODE2 0X4C
+#define P0_WIN_MODE 0X50
+#define P0_OUT_WIN_Y1_HIGH 0X51
+#define P0_OUT_WIN_Y1_LOW 0X52
+#define P0_OUT_WIN_X1_HIGH 0X53
+#define P0_OUT_WIN_X1_LOW 0X54
+#define P0_OUT_WIN_HEIGHT_HIGH 0X55
+#define P0_OUT_WIN_HEIGHT_LOW 0X56
+#define P0_OUT_WIN_WIDTH_HIGH 0X57
+#define P0_OUT_WIN_WIDTH_LOW 0X58
+
+#define P0_GLOBAL_SATURATION 0XD0
+#define P0_SATURATION_CB 0XD1
+#define P0_SATURATION_CR 0XD2
+#define P0_LUMA_CONTRAST 0XD3
+#define P0_CONTRAST_CENTER 0XD4
+#define P0_LUMA_OFFSET 0XD5
+#define P0_FIXED_CB 0XDA
+#define P0_FIXED_CR 0XDB
+
+//----page3-----------------------------
+#define P3_IMAGE_WIDTH_LOW 0X5B
+#define P3_IMAGE_WIDTH_HIGH 0X5C
+#define P3_IMAGE_HEIGHT_LOW 0X5D
+#define P3_IMAGE_HEIGHT_HIGH 0X5E
+
+
+#endif //__GC032A_REG_REGS_H__
--- a/code/components/esp32-camera-master/sensors/private_include/gc032a_settings.h
+++ b/code/components/esp32-camera-master/sensors/private_include/gc032a_settings.h
@@ -0,0 +1,401 @@
+#ifndef _GC032A_SETTINGS_H_
+#define _GC032A_SETTINGS_H_
+
+#include <stdint.h>
+#include <stdbool.h>
+#include "esp_attr.h"
+#include "gc032a_regs.h"
+
+
+#define REG_DLY 0xffff
+#define REGLIST_TAIL 0x0000
+
+
+/*
+ * The default register settings, as obtained from OmniVision.  There
+ * is really no making sense of most of these - lots of "reserved" values
+ * and such.
+ *
+ */
+static const uint16_t gc032a_default_regs[][2] = {
+    /*System*/
+    {0xf3, 0xff},
+    {0xf5, 0x06},
+    {0xf7, 0x01},
+    {0xf8, 0x03},
+    {0xf9, 0xce},
+    {0xfa, 0x00},
+    {0xfc, 0x02},
+    {0xfe, 0x02},
+    {0x81, 0x03},
+
+    {0xfe, 0x00},
+    {0x77, 0x64},
+    {0x78, 0x40},
+    {0x79, 0x60},
+    /*ANALOG & CISCTL*/
+    {0xfe, 0x00},
+    {0x03, 0x01},
+    {0x04, 0xce},
+    {0x05, 0x01},
+    {0x06, 0xad},
+    {0x07, 0x00},
+    {0x08, 0x10},
+    {0x0a, 0x00},
+    {0x0c, 0x00},
+    {0x0d, 0x01},
+    {0x0e, 0xe8}, // height 488
+    {0x0f, 0x02},
+    {0x10, 0x88}, // width 648
+    {0x17, 0x54},
+    {0x19, 0x08},
+    {0x1a, 0x0a},
+    {0x1f, 0x40},
+    {0x20, 0x30},
+    {0x2e, 0x80},
+    {0x2f, 0x2b},
+    {0x30, 0x1a},
+    {0xfe, 0x02},
+    {0x03, 0x02},
+    {0x05, 0xd7},
+    {0x06, 0x60},
+    {0x08, 0x80},
+    {0x12, 0x89},
+
+    /*blk*/
+    {0xfe, 0x00},
+    {0x18, 0x02},
+    {0xfe, 0x02},
+    {0x40, 0x22},
+    {0x45, 0x00},
+    {0x46, 0x00},
+    {0x49, 0x20},
+    {0x4b, 0x3c},
+    {0x50, 0x20},
+    {0x42, 0x10},
+
+    /*isp*/
+    {0xfe, 0x01},
+    {0x0a, 0xc5},
+    {0x45, 0x00},
+    {0xfe, 0x00},
+    {0x40, 0xff},
+    {0x41, 0x25},
+    {0x42, 0xcf},
+    {0x43, 0x10},
+    {0x44, 0x83},
+    {0x46, 0x23},
+    {0x49, 0x03},
+    {0x52, 0x02},
+    {0x54, 0x00},
+    {0xfe, 0x02},
+    {0x22, 0xf6},
+
+    /*Shading*/
+    {0xfe, 0x01},
+    {0xc1, 0x38},
+    {0xc2, 0x4c},
+    {0xc3, 0x00},
+    {0xc4, 0x32},
+    {0xc5, 0x24},
+    {0xc6, 0x16},
+    {0xc7, 0x08},
+    {0xc8, 0x08},
+    {0xc9, 0x00},
+    {0xca, 0x20},
+    {0xdc, 0x8a},
+    {0xdd, 0xa0},
+    {0xde, 0xa6},
+    {0xdf, 0x75},
+
+    /*AWB*/
+    {0xfe, 0x01},
+    {0x7c, 0x09},
+    {0x65, 0x06},
+    {0x7c, 0x08},
+    {0x56, 0xf4},
+    {0x66, 0x0f},
+    {0x67, 0x84},
+    {0x6b, 0x80},
+    {0x6d, 0x12},
+    {0x6e, 0xb0},
+    {0x86, 0x00},
+    {0x87, 0x00},
+    {0x88, 0x00},
+    {0x89, 0x00},
+    {0x8a, 0x00},
+    {0x8b, 0x00},
+    {0x8c, 0x00},
+    {0x8d, 0x00},
+    {0x8e, 0x00},
+    {0x8f, 0x00},
+    {0x90, 0x00},
+    {0x91, 0x00},
+    {0x92, 0xf4},
+    {0x93, 0xd5},
+    {0x94, 0x50},
+    {0x95, 0x0f},
+    {0x96, 0xf4},
+    {0x97, 0x2d},
+    {0x98, 0x0f},
+    {0x99, 0xa6},
+    {0x9a, 0x2d},
+    {0x9b, 0x0f},
+    {0x9c, 0x59},
+    {0x9d, 0x2d},
+    {0x9e, 0xaa},
+    {0x9f, 0x67},
+    {0xa0, 0x59},
+    {0xa1, 0x00},
+    {0xa2, 0x00},
+    {0xa3, 0x0a},
+    {0xa4, 0x00},
+    {0xa5, 0x00},
+    {0xa6, 0xd4},
+    {0xa7, 0x9f},
+    {0xa8, 0x55},
+    {0xa9, 0xd4},
+    {0xaa, 0x9f},
+    {0xab, 0xac},
+    {0xac, 0x9f},
+    {0xad, 0x55},
+    {0xae, 0xd4},
+    {0xaf, 0xac},
+    {0xb0, 0xd4},
+    {0xb1, 0xa3},
+    {0xb2, 0x55},
+    {0xb3, 0xd4},
+    {0xb4, 0xac},
+    {0xb5, 0x00},
+    {0xb6, 0x00},
+    {0xb7, 0x05},
+    {0xb8, 0xd6},
+    {0xb9, 0x8c},
+
+    /*CC*/
+    {0xfe, 0x01},
+    {0xd0, 0x40},
+    {0xd1, 0xf8},
+    {0xd2, 0x00},
+    {0xd3, 0xfa},
+    {0xd4, 0x45},
+    {0xd5, 0x02},
+
+    {0xd6, 0x30},
+    {0xd7, 0xfa},
+    {0xd8, 0x08},
+    {0xd9, 0x08},
+    {0xda, 0x58},
+    {0xdb, 0x02},
+    {0xfe, 0x00},
+
+    /*Gamma*/
+    {0xfe, 0x00},
+    {0xba, 0x00},
+    {0xbb, 0x04},
+    {0xbc, 0x0a},
+    {0xbd, 0x0e},
+    {0xbe, 0x22},
+    {0xbf, 0x30},
+    {0xc0, 0x3d},
+    {0xc1, 0x4a},
+    {0xc2, 0x5d},
+    {0xc3, 0x6b},
+    {0xc4, 0x7a},
+    {0xc5, 0x85},
+    {0xc6, 0x90},
+    {0xc7, 0xa5},
+    {0xc8, 0xb5},
+    {0xc9, 0xc2},
+    {0xca, 0xcc},
+    {0xcb, 0xd5},
+    {0xcc, 0xde},
+    {0xcd, 0xea},
+    {0xce, 0xf5},
+    {0xcf, 0xff},
+
+    /*Auto Gamma*/
+    {0xfe, 0x00},
+    {0x5a, 0x08},
+    {0x5b, 0x0f},
+    {0x5c, 0x15},
+    {0x5d, 0x1c},
+    {0x5e, 0x28},
+    {0x5f, 0x36},
+    {0x60, 0x45},
+    {0x61, 0x51},
+    {0x62, 0x6a},
+    {0x63, 0x7d},
+    {0x64, 0x8d},
+    {0x65, 0x98},
+    {0x66, 0xa2},
+    {0x67, 0xb5},
+    {0x68, 0xc3},
+    {0x69, 0xcd},
+    {0x6a, 0xd4},
+    {0x6b, 0xdc},
+    {0x6c, 0xe3},
+    {0x6d, 0xf0},
+    {0x6e, 0xf9},
+    {0x6f, 0xff},
+
+    /*Gain*/
+    {0xfe, 0x00},
+    {0x70, 0x50},
+
+    /*AEC*/
+    {0xfe, 0x00},
+    {0x4f, 0x01},
+    {0xfe, 0x01},
+    {0x0d, 0x00},
+    {0x12, 0xa0},
+    {0x13, 0x3a},
+    {0x44, 0x04},
+    {0x1f, 0x30},
+    {0x20, 0x40},
+    {0x26, 0x9a},
+    {0x3e, 0x20},
+    {0x3f, 0x2d},
+    {0x40, 0x40},
+    {0x41, 0x5b},
+    {0x42, 0x82},
+    {0x43, 0xb7},
+    {0x04, 0x0a},
+    {0x02, 0x79},
+    {0x03, 0xc0},
+
+    /*measure window*/
+    {0xfe, 0x01},
+    {0xcc, 0x08},
+    {0xcd, 0x08},
+    {0xce, 0xa4},
+    {0xcf, 0xec},
+
+    /*DNDD*/
+    {0xfe, 0x00},
+    {0x81, 0xb8},
+    {0x82, 0x12},
+    {0x83, 0x0a},
+    {0x84, 0x01},
+    {0x86, 0x50},
+    {0x87, 0x18},
+    {0x88, 0x10},
+    {0x89, 0x70},
+    {0x8a, 0x20},
+    {0x8b, 0x10},
+    {0x8c, 0x08},
+    {0x8d, 0x0a},
+
+    /*Intpee*/
+    {0xfe, 0x00},
+    {0x8f, 0xaa},
+    {0x90, 0x9c},
+    {0x91, 0x52},
+    {0x92, 0x03},
+    {0x93, 0x03},
+    {0x94, 0x08},
+    {0x95, 0x44},
+    {0x97, 0x00},
+    {0x98, 0x00},
+
+    /*ASDE*/
+    {0xfe, 0x00},
+    {0xa1, 0x30},
+    {0xa2, 0x41},
+    {0xa4, 0x30},
+    {0xa5, 0x20},
+    {0xaa, 0x30},
+    {0xac, 0x32},
+
+    /*YCP*/
+    {0xfe, 0x00},
+    {0xd1, 0x3c},
+    {0xd2, 0x3c},
+    {0xd3, 0x38},
+    {0xd6, 0xf4},
+    {0xd7, 0x1d},
+    {0xdd, 0x73},
+    {0xde, 0x84},
+
+    /*Banding*/
+    {0xfe, 0x00},
+    {0x05, 0x01},
+    {0x06, 0xad},
+    {0x07, 0x00},
+    {0x08, 0x10},
+
+    {0xfe, 0x01},
+    {0x25, 0x00},
+    {0x26, 0x9a},
+
+    {0x27, 0x01},
+    {0x28, 0xce},
+    {0x29, 0x02},
+    {0x2a, 0x68},
+    {0x2b, 0x02},
+    {0x2c, 0x68},
+    {0x2d, 0x07},
+    {0x2e, 0xd2},
+    {0x2f, 0x0b},
+    {0x30, 0x6e},
+    {0x31, 0x0e},
+    {0x32, 0x70},
+    {0x33, 0x12},
+    {0x34, 0x0c},
+    {0x3c, 0x30},
+
+    /*Analog&Cisctl*/
+    {0xfe, 0x00},
+    {0x05, 0x01},
+    {0x06, 0xa0},
+    {0x07, 0x00},
+    {0x08, 0x20},
+    {0x0a, 0x78},
+    {0x0c, 0xa0},
+    {0x0d, 0x00}, //window_height [8]
+    {0x0e, 0xf8}, //window_height [7:0] 248
+    {0x0f, 0x01}, //window_width [9:8]
+    {0x10, 0x48}, //window_width [7:0]  328
+
+    {0x55, 0x00},
+    {0x56, 0xf0}, // 240
+    {0x57, 0x01},
+    {0x58, 0x40}, // 320
+
+    /*SPI*/
+    {0xfe, 0x03},
+    {0x5b, 0x40},
+    {0x5c, 0x01},
+    {0x5d, 0xf0},
+    {0x5e, 0x00},
+
+    /*AEC*/
+    {0xfe, 0x01},
+    {0x25, 0x00}, //step
+    {0x26, 0x63},
+    {0x27, 0x01},
+    {0x28, 0x29},
+    {0x29, 0x01},
+    {0x2a, 0x29},
+    {0x2b, 0x01},
+    {0x2c, 0x29},
+    {0x2d, 0x01},
+    {0x2e, 0x29},
+    {0x2f, 0x01},
+    {0x30, 0x29},
+    {0x31, 0x01},
+    {0x32, 0x29},
+    {0x33, 0x01},
+    {0x34, 0x29},
+    {0x3c, 0x00},
+
+    /*measure window*/
+    {0xfe, 0x01},
+    {0xcc, 0x04},
+    {0xcd, 0x04},
+    {0xce, 0x72},
+    {0xcf, 0x52},
+    {REGLIST_TAIL, 0x00},
+};
+
+#endif
--- a/code/components/esp32-camera-master/sensors/private_include/gc2145.h
+++ b/code/components/esp32-camera-master/sensors/private_include/gc2145.h
@@ -0,0 +1,27 @@
+
+#ifndef __GC2145_H__
+#define __GC2145_H__
+
+#include "sensor.h"
+
+/**
+ * @brief Detect sensor pid
+ *
+ * @param slv_addr SCCB address
+ * @param id Detection result
+ * @return
+ *     0:       Can't detect this sensor
+ *     Nonzero: This sensor has been detected
+ */
+int gc2145_detect(int slv_addr, sensor_id_t *id);
+
+/**
+ * @brief initialize sensor function pointers
+ *
+ * @param sensor pointer of sensor
+ * @return
+ *      Always 0
+ */
+int gc2145_init(sensor_t *sensor);
+
+#endif // __GC2145_H__
--- a/Show More
+++ b/Show More
Author	SHA1	Message	Date
jomjol	aca60465f0	v11.1.1	2022-08-22 18:20:00 +02:00
jomjol	57bdca37fc	v11.1.1	2022-08-22 18:12:08 +02:00
jomjol	6409397770	Merge pull request #933 from haverland/rolling Fix for #921, #919	2022-08-22 17:58:30 +02:00
Frank Haverland	974044adf0	last analog result_float as Readout parameter, testcases for #921 , #919	2022-08-22 16:10:07 +02:00
Frank Haverland	59aeeda786	Merge branch 'rolling' into analogtodig_as_float	2022-08-22 10:33:10 +02:00
Frank Haverland	b6bf8d992f	Test case for postprocessing	2022-08-22 10:31:25 +02:00
jomjol	9d31edc67a	Rolling 2022-08-21	2022-08-21 21:33:56 +02:00
jomjol	c9a879d329	v11.1.0	2022-08-21 17:56:46 +02:00
jomjol	ea69b1be00	v11.1.0	2022-08-21 17:44:34 +02:00
jomjol	2a8b3a87ea	Merge pull request #922 from haverland/rolling Rolling	2022-08-21 16:47:07 +02:00
jomjol	52783734ce	Merge pull request #925 from jochenchrist/master Remove .DS_Store	2022-08-21 16:44:12 +02:00
jomjol	0e1b390ec6	Merge pull request #918 from ppisljar/patch-1 Update FeatureRequest.md	2022-08-21 16:41:30 +02:00
jochen	ab49bdf82f	.DS_Store removed	2022-08-20 19:25:52 +02:00
jochenchrist	25e7051271	Delete .DS_Store	2022-08-20 19:23:08 +02:00
Frank Haverland	7315f9adfc	Merge branch 'jomjol:rolling' into rolling	2022-08-19 21:33:08 +02:00
Frank Haverland	af1aee4ac3	add testcase for #921	2022-08-19 21:30:11 +02:00
Frank Haverland	d6ff7eef88	fix problems with early transition of digits if analog pointers. #921	2022-08-19 21:05:23 +02:00
Frank Haverland	7706b4dbc3	fix for #919 the prev is int, so <9.0 instead of <9.5	2022-08-18 19:28:13 +02:00
Peter Pisljar	3561ecd2b7	Update FeatureRequest.md	2022-08-18 15:16:57 +02:00
jomjol	74c7ff7fdf	v11.0.1	2022-08-15 22:48:42 +02:00
jomjol	a68ce353ad	Merge pull request #910 from haverland/rolling Fix naming of models and new version	2022-08-13 15:58:57 +02:00
Frank Haverland	0d168f3445	Merge branch 'jomjol:rolling' into rolling	2022-08-13 15:38:57 +02:00
Frank Haverland	073e04a3cc	fix naming of models and new versions	2022-08-13 15:37:04 +02:00
jomjol	591dc048d4	v11.0.0	2022-08-13 14:26:04 +02:00
jomjol	bfe8d3b37a	v11.0.0	2022-08-13 14:20:40 +02:00
jomjol	9695dba415	Merge branch 'master' into rolling	2022-08-07 21:20:28 +02:00
jomjol	6a48f0502e	Merge pull request #885 from haverland/rolling CNNThreshold removed vor Analog100 and Digital100	2022-08-07 21:19:37 +02:00
Frank Haverland	4a8d6592ab	CNNThreshold removed for Analog100 and Digital100	2022-07-28 19:43:45 +02:00
jomjol	434aebd641	Merge pull request #881 from haverland/rolling Ignore hidden files in configuration->model selection	2022-07-25 19:11:29 +02:00
Frank Haverland	c124c38e70	ignore hidden files in configuration->model selection	2022-07-25 16:30:11 +02:00
Frank Haverland	e6d60bb124	Merge branch 'jomjol:rolling' into rolling	2022-07-24 20:20:53 +02:00
jomjol	085ea2028c	v10.6.1	2022-07-24 19:07:43 +02:00
jomjol	0e7c600cf7	Rolling v10.6.1	2022-07-24 18:53:25 +02:00
Frank Haverland	3f3532defe	Revert "Fix for #712 "Incorrect rollover digital numbers"" This reverts commit `11bfaf0e91`.	2022-07-20 19:12:19 +02:00
Frank Haverland	a0ffc88e47	Merge branch 'rolling' of https://github.com/haverland/AI-on-the-edge-device into rolling	2022-07-20 18:37:05 +02:00
Frank Haverland	11bfaf0e91	Fix for #712 "Incorrect rollover digital numbers"	2022-07-20 18:35:42 +02:00
jomjol	189093d548	Merge pull request #862 from haverland/rolling Version 1.0 of digital and analog categorical models	2022-07-18 20:33:02 +02:00
Frank Haverland	34eb89b1b6	fix extension finding for tlfite modellist	2022-07-18 19:18:09 +02:00
Frank Haverland	b725d242d3	Add Error to Logfile if output-dimenstion of model inconsistent.	2022-07-18 18:36:42 +02:00
Frank Haverland	568e88314b	Merge branch 'jomjol:rolling' into rolling	2022-07-17 20:00:28 +02:00
jomjol	aaad952782	v10.6.0	2022-07-17 09:34:27 +02:00
jomjol	dc27911174	Preparation for v10.6.0	2022-07-17 09:00:10 +02:00
Frank Haverland	42678ae8e1	Merge branch 'jomjol:rolling' into rolling	2022-07-16 23:09:41 +02:00
Frank Haverland	b6341992f6	Version 1.0 of digital and analog categorical models	2022-07-16 21:29:56 +02:00
jomjol	17fd0f96df	Rolling 20220716_2	2022-07-16 20:59:09 +02:00
jomjol	0b039e8d8c	Merge pull request #861 from haverland/rolling fix false value if analog + dighybrid on last digit if previous >=9.5	2022-07-16 20:07:02 +02:00
Frank Haverland	eab8a6d96b	fix false value if analog + dighybrid on last digit if previous >=9.5	2022-07-16 19:37:30 +02:00
jomjol	9f72bf117e	Rolling 20220716	2022-07-16 08:09:18 +02:00
jomjol	058e9436f0	Merge pull request #858 from haverland/rolling Rolling: Add analog classification model	2022-07-16 07:50:00 +02:00
Frank Haverland	ebd8fe0e4f	Merge branch 'rolling' of https://github.com/haverland/AI-on-the-edge-device into rolling	2022-07-03 21:02:00 +02:00
Frank Haverland	7970f5f691	Add analog classification (100) as modeltype analog100	2022-07-03 21:01:42 +02:00
jomjol	0689ca86c5	Merge branch 'rolling' of https://github.com/jomjol/AI-on-the-edge-device into rolling	2022-07-01 07:27:28 +02:00
jomjol	5783e0f182	Rolling 20220107	2022-07-01 07:27:26 +02:00
jomjol	e190aa8d03	Merge pull request #839 from haverland/rolling Rolling: Add testing for CNNFlowcontroll	2022-06-28 07:09:35 +02:00
Frank Haverland	15bac02cbf	Merge branch 'jomjol:rolling' into rolling	2022-06-26 23:23:16 +02:00
Frank Haverland	7fdacce9a2	Merge branch 'rolling' of https://github.com/haverland/AI-on-the-edge-device into rolling	2022-06-26 23:21:39 +02:00
Frank Haverland	78ea179e9a	added testing to cnnflowcontroll::ZeigerEval/ZeigerEvalHybrid	2022-06-26 23:21:01 +02:00
jomjol	2cfc3fadb2	Rolling 20220626	2022-06-26 21:57:20 +02:00
jomjol	60cdee2ea6	Merge pull request #834 from haverland/rolling Rolling: Fix -1 issue on watermeter (digital100 + analog)	2022-06-26 21:12:43 +02:00
Frank Haverland	a17ac6860d	Merge branch 'jomjol:rolling' into rolling	2022-06-26 13:47:06 +02:00
Frank Haverland	ef24466702	revert comment out in ZeigerEval use ZeigerEvalHybrid instead of ZeigerEval in DoubleHyprid10/Digital100 branch	2022-06-26 13:44:55 +02:00
jomjol	de78027b0e	Merge pull request #837 from toolsfactory/master Added Homie Mqtt request	2022-06-26 12:11:35 +02:00
jomjol	28c40a0ad2	Merge branch 'rolling' into master	2022-06-26 12:11:02 +02:00
Michael Geissler	230bbb2239	Added Homie Mqtt request	2022-06-26 08:14:41 +02:00
Frank Haverland	bda2913a32	cleanup, disable debugging	2022-06-24 14:59:51 +02:00
Frank Haverland	78daaf6e5b	fix -1 on the last digital digit if using digital100 model	2022-06-24 14:58:00 +02:00
jomjol	8939167a39	Rolling 20220618_v3	2022-06-18 16:01:25 +02:00
jomjol	b48c6111d9	Merge pull request #827 from haverland/rolling Rolling: Add hostname to title of configuration	2022-06-18 16:00:24 +02:00
Frank Haverland	de1e919b96	add hostname to configure	2022-06-18 15:39:54 +02:00
jomjol	35a96027f1	Rolling 20220618_v2	2022-06-18 14:42:30 +02:00
jomjol	7a36bfa2ff	Rolling 20220618	2022-06-18 11:10:31 +02:00
jomjol	dfeac0cb7f	Rolling 20220609	2022-06-09 21:16:05 +02:00
jomjol	dfec780157	Merge pull request #817 from haverland/rolling Digital100 model processing added	2022-06-01 20:34:43 +02:00
Frank Haverland	4cc93324f8	Digital100 model	2022-05-31 22:33:12 +02:00
Frank Haverland	bf8833eae6	add Digital100 model processing with output 0.0-9.9 as categorical network	2022-05-30 23:22:17 +02:00
jomjol	00028010ee	Rolling 20220526	2022-05-26 20:31:26 +02:00
jomjol	cce812ff11	Rolling 20220417	2022-04-17 21:51:26 +02:00
jomjol	ccb4bd595c	Rolling 20220415	2022-04-15 14:49:58 +02:00
jomjol	dc7eedcd8f	Merge pull request #774 from wetneb/534-influxdb InfluxDB integration	2022-04-15 14:39:53 +02:00
jomjol	40faa78929	Merge branch 'rolling' into 534-influxdb	2022-04-15 14:39:09 +02:00
Antonin Delpeuch	eb53db00d0	First draft of InfluxDB integration, for #534	2022-04-15 10:19:14 +02:00
jomjol	658ef39736	Rolling 20220322	2022-03-22 20:51:55 +01:00
jomjol	0e90bcb2ef	Rolling 20220320	2022-03-20 21:36:44 +01:00
jomjol	a020fce2d7	Merge pull request #725 from mkelley88/master Update README.md	2022-03-15 20:27:34 +01:00
Matthew T. Kelley	525ccf7aba	Update README.md Made changes to increase clarity of documentation.	2022-03-15 04:50:37 -04:00
jomjol	ebcfc16f63	Create dig-s0-q-20220224.tflite	2022-02-24 21:17:54 +01:00
jomjol	4b825efffb	v10.5.2	2022-02-22 19:17:20 +01:00
jomjol	71871016d2	v10.5.1	2022-02-20 16:24:05 +01:00
jomjol	c07ef23d5b	v10.5.1	2022-02-20 16:05:36 +01:00
jomjol	cbe884ad63	v10.5.0	2022-02-18 21:22:00 +01:00
jomjol	1dd703b337	v10.5.0	2022-02-18 21:14:04 +01:00
jomjol	bcb1d98191	Rolling 20220215	2022-02-15 21:29:09 +01:00
jomjol	1f5486e8cc	Rolling 20220215	2022-02-15 21:26:38 +01:00
jomjol	1371be6f2c	v10.4.0	2022-02-12 09:11:43 +01:00
jomjol	b0d8ed6248	v10.4.0	2022-02-12 09:02:00 +01:00
jomjol	379f4585e3	Rolling 20220211	2022-02-11 21:38:40 +01:00
jomjol	45a71981c8	Rolling 20220208	2022-02-08 19:36:25 +01:00
jomjol	ac3409f644	Update FeatureRequest.md	2022-02-06 20:24:59 +01:00
jomjol	11c33f81dd	Rolling 20220206	2022-02-06 19:50:47 +01:00
jomjol	641cc860d8	v10.3.0	2022-01-29 15:59:13 +01:00
jomjol	2029bd6e8a	Rolling 20220122	2022-01-29 15:41:56 +01:00
jomjol	1ca5e1218d	Rolling 20220128	2022-01-28 19:17:05 +01:00
jomjol	887c704f63	Rolling 20220127	2022-01-27 21:43:54 +01:00
jomjol	567dc74cd7	Rolling 20220123	2022-01-23 19:47:34 +01:00
jomjol	53606d5055	Rolling 20220121	2022-01-21 20:50:54 +01:00
jomjol	19a6c21c44	Rolling 20220118	2022-01-18 07:16:38 +01:00
jomjol	3a4b11e4b3	v10.2.0	2022-01-14 21:03:38 +01:00
jomjol	d981574ab2	Update FeatureRequest.md	2022-01-14 20:42:24 +01:00
jomjol	eb817739b9	v10.2.0	2022-01-14 20:34:59 +01:00
jomjol	8972f17638	Merge branch 'wifi-mod'	2022-01-14 20:28:42 +01:00
jomjol	4d997d9473	v10.2.0	2022-01-14 20:28:01 +01:00
jomjol	e79c86c7b6	v10.2.0	2022-01-14 20:18:24 +01:00
jomjol	96bb86536f	v10.1.1	2022-01-12 21:02:05 +01:00
jomjol	2daf6c8b3f	v10.1.0	2022-01-09 09:47:16 +01:00
jomjol	24b46158de	v10.1.0	2022-01-09 09:43:22 +01:00
jomjol	63d336ba28	Rolling 20220104	2022-01-04 21:17:20 +01:00
jomjol	8dd3a92671	Rolling 20220103	2022-01-03 17:32:01 +01:00
jomjol	8e8897c70d	v10.0.2	2022-01-01 08:57:07 +01:00
jomjol	eac513411e	v10.0.2	2022-01-01 08:53:55 +01:00
jomjol	98f9274085	V10.0.1	2021-12-31 09:02:59 +01:00
jomjol	884dd9fc3b	V10.0.0	2021-12-30 18:38:15 +01:00
jomjol	04c564936a	Master 10.0.0	2021-12-30 18:29:01 +01:00
jomjol	957138960a	Rolling 20211230	2021-12-30 11:02:20 +01:00
jomjol	58a0297915	Rolling 20211223	2021-12-23 08:03:46 +01:00
jomjol	61bf536207	Rolling 20211212	2021-12-12 19:30:07 +01:00
jomjol	4136a7b372	Merge pull request #456 from mad2xlc/master tabindex to html	2021-12-12 17:45:25 +01:00
Stefan	b3afc9961d	Merge branch 'jomjol:master' into master	2021-12-12 10:36:53 +01:00
jomjol	27e2e042da	Rolling 2021-21-03	2021-12-03 18:12:45 +01:00
Frederik Kemner	cc659bad30	Time based flow rate limiting Take time since last valid value into account for flow rate limiting	2021-12-03 11:27:20 +01:00
jomjol	776931c0ad	v9.2.0	2021-12-02 21:56:05 +01:00
jomjol	e22b4b6fe1	v9.2.0	2021-12-02 21:52:45 +01:00
jomjol	cad534bffe	update wiki	2021-12-02 21:47:37 +01:00
jomjol	3b44adb6fa	Rolling 20211128	2021-11-28 09:09:24 +01:00
jomjol	cc0bd1473f	Rolling 20211124	2021-11-24 07:32:03 +01:00
jomjol	58124d27bf	Rolling 2021-11-23	2021-11-23 17:57:07 +01:00
Sven Rojek	9ad118814a	add basic exception handling for the camera module	2021-11-21 19:05:17 +01:00
jomjol	270f8dd093	v9.1.1	2021-11-16 07:15:55 +01:00
jomjol	fde0ae4781	v9.1.0	2021-11-14 08:57:19 +01:00
jomjol	b87ce8c75d	Merge branch 'master' into rolling	2021-11-14 08:52:40 +01:00
jomjol	e372c87836	v9.1.0	2021-11-14 08:52:28 +01:00
jomjol	f8478d7742	Merge branch 'master' into rolling	2021-11-14 08:48:42 +01:00
jomjol	a3d6923fec	v9.1.0	2021-11-14 08:47:55 +01:00
jomjol	7bfa22187d	Merge branch 'pr/398' into rolling	2021-11-14 08:19:13 +01:00
pixel::doc	7a8affae32	Update Helper.cpp Change "open config file" to "open file"	2021-11-14 01:05:40 +01:00
jomjol	e48dd7c4c4	rolling 20211106	2021-11-06 22:46:16 +01:00
mad2xlc	6d298c1746	added tabindex for coordinate fields	2021-11-04 17:17:22 +01:00
jomjol	4fe9ab92e0	image update	2021-10-27 18:57:07 +02:00
jomjol	f2ac4cdc64	v9.0.0	2021-10-23 16:35:40 +02:00
jomjol	be902401d1	Merge branch 'rolling'	2021-10-23 16:32:44 +02:00
jomjol	020e93bca2	v9.0.0	2021-10-23 16:31:55 +02:00
jomjol	61dfdc2258	Add files via upload	2021-10-23 15:44:46 +02:00
jomjol	a98e3c8eab	Add files via upload	2021-10-22 21:31:21 +02:00
jomjol	58a90ff706	v8.5.0	2021-10-07 07:45:40 +02:00
jomjol	d0bf12f3d4	v8.5.0	2021-10-07 07:16:46 +02:00
jomjol	af16785bbf	Rolling 20211002	2021-10-02 14:59:09 +02:00
jomjol	18f6e83a2c	v8.4.0	2021-09-25 18:57:40 +02:00
jomjol	147d97421b	Merge branch 'rolling'	2021-09-25 18:53:47 +02:00
jomjol	dcf2feb7aa	v8.4.0	2021-09-25 18:53:14 +02:00
jomjol	e63e940b96	v8.4.0	2021-09-25 08:08:21 +02:00
jomjol	68b0fb83ee	v8.4.0	2021-09-24 19:57:48 +02:00
jomjol	f15e5f060a	v8.4.0	2021-09-23 18:43:53 +02:00
jomjol	e2a403441f	Rolling 20210922	2021-09-22 22:13:08 +02:00
jomjol	9b3665b9c6	Rolling 20210921 v2	2021-09-21 19:41:20 +02:00
jomjol	f4c8bf9206	Rolling 20210921	2021-09-21 18:49:32 +02:00
jomjol	c033db9c31	Rolling 20210921	2021-09-21 07:27:46 +02:00
jomjol	9300526f49	Rolling 2021-09-20	2021-09-20 21:18:34 +02:00
jomjol	b6dd1f7f2d	Update	2021-09-14 20:00:45 +02:00
jomjol	1e6eddca04	Rolling 20210913	2021-09-13 20:05:54 +02:00
jomjol	19ca0d7dd7	Update Versioninfo	2021-09-12 07:33:30 +02:00
jomjol	7fcb5d1c0c	v8.3.0	2021-09-12 07:29:30 +02:00
jomjol	dd995ec28a	Rolling 20210910	2021-09-10 09:26:52 +02:00
jomjol	af99de3535	IgnoreLeadingNaN	2021-09-02 11:04:01 +02:00
jomjol	3567cc2fb0	Merge pull request #330 from pixeldoc2000/pixeldoc2000-patch-1 Pixeldoc2000 patch 1	2021-09-02 11:00:49 +02:00
pixel::doc	5e9d9bd264	Update edit_config_param.html Fixed some more Text.	2021-09-02 10:09:36 +02:00
pixel::doc	62447c1bb9	Update edit_config_param.html Fixed some Text	2021-09-02 00:23:59 +02:00
jomjol	a86434c9a2	Rolling 20210831	2021-08-31 11:40:29 +02:00
jomjol	b7b70299f7	Rolling 20210830	2021-08-30 21:21:18 +02:00
jomjol	eb02e0aec1	new images	2021-08-29 20:57:21 +02:00
jomjol	7816e53db7	v8.2.0	2021-08-24 08:37:18 +02:00
jomjol	7ae08e572a	Merge branch 'rolling'	2021-08-24 08:34:32 +02:00
jomjol	47d15d8adb	v8.2.0	2021-08-24 08:33:56 +02:00
jomjol	0dac0e87e4	rolling 20210823	2021-08-23 18:57:48 +02:00
jomjol	b290099d5b	v12.0.0	2021-08-12 07:25:12 +02:00
jomjol	f6b1a41a0b	v12.0.0	2021-08-12 07:20:58 +02:00
jomjol	e529af04cf	Update FeatureRequest.md	2021-08-10 20:19:05 +02:00
jomjol	6c365dd949	rolling v20210809	2021-08-09 21:53:07 +02:00
jomjol	32f15fc557	rolling 20210708	2021-08-07 15:25:27 +02:00
jomjol	6f06af1d5f	Update README.md	2021-08-01 21:52:00 +02:00