andersontruong / GroundSight

Context-based interaction using embedded deep learning

TensorFlow Lite Micro Setup with PlatformIO (ESP32-CAM)

This setup guide unifies several sources to cover the entire pipeline. We will be using the ESP32-CAM board for demonstration, but most of the steps should work for many MCUs.

Installations: VSCode, PlatformIO extension, Python (conda, TensorFlow-GPU (cuda, cudnn), Matplotlib, NumPy, JupyterLab), Synthesis (xxd for Linux or Python script), tflite-micro library, tflm class for setup and utilities)

Setup

1. VSCode and PlatformIO extension
2. Python and TensorFlow-GPU setup
3. Setup synthesis method
4. Download tflite-micro library
5. Download TFLM helper class (optional)

Sections

1. Training
2. Quantization
3. Synthesis
4. Deployment