espressif/esp-tflite-micro

What it solves

This project provides a way to run TensorFlow Lite Micro (TFLite Micro) on Espressif chipsets (such as the ESP32-P4, ESP32-S3, and ESP32-C3) using the ESP-IDF platform. It enables the deployment of small, optimized machine learning models on resource-constrained microcontroller hardware.

How it works

It acts as a component for the ESP-IDF ecosystem, integrating the TFLite Micro library. To improve performance, it integrates with ESP-NN, a library of optimized kernel implementations specifically designed for Espressif hardware, which significantly reduces the time required for model execution (inference).

Who it’s for

Developers working with Espressif microcontrollers who want to implement on-device AI, such as person detection or speech recognition, directly on the hardware.

Highlights

• Hardware Optimization: Uses ESP-NN to drastically speed up inference (e.g., reducing person detection time from 2300ms to 54ms on ESP32-S3).
• Ready-to-use Examples: Includes pre-built examples for person detection, micro speech, and a basic hello world demo.
• ESP-IDF Integration: Easily installable as a dependency via idf.py add-dependency.
• C-based Implementation: Based on the upstream TensorFlow Lite Micro repository for minimal memory footprint.