ESP-NN or ESP-DL for ESP32-S3
Posted: Fri Sep 13, 2024 7:03 am
Hello everyone! I am new here and now trying to formulate a plan, about running a customized AI model(for example, the AI model generated by using tensorflow / scikit-learn) on ESP32-S3. I have some questions related to this:
1. I think I should use ESP-NN. But there is also a tool called ESP-DL. What's the differences between them and, any limitations by using those frameworks?
2. I think I should use ESP-PPQ for quantizing the model first and then using TVM tool to deploy. But all of those tools are in "ESP-DL" ecosystem: https://docs.espressif.com/projects/esp ... l#overview. From ESP32-S3 product page it says ESP-NN should be the AI framework used so, how should I move forward?
3. I will need also FreeRTOS as scheduler, executing on the same processor as where the AI engine is running. Is that possible based on the current ESP-IDF capability?
4. There is an interesting project "Tensorflow Lite for Microcontroller". Can I actually run this with ESP-IDF currently or any tutorials I could take a look?
Many, many thanks in advance and best regards!
1. I think I should use ESP-NN. But there is also a tool called ESP-DL. What's the differences between them and, any limitations by using those frameworks?
2. I think I should use ESP-PPQ for quantizing the model first and then using TVM tool to deploy. But all of those tools are in "ESP-DL" ecosystem: https://docs.espressif.com/projects/esp ... l#overview. From ESP32-S3 product page it says ESP-NN should be the AI framework used so, how should I move forward?
3. I will need also FreeRTOS as scheduler, executing on the same processor as where the AI engine is running. Is that possible based on the current ESP-IDF capability?
4. There is an interesting project "Tensorflow Lite for Microcontroller". Can I actually run this with ESP-IDF currently or any tutorials I could take a look?
Many, many thanks in advance and best regards!