Tensorflow Micro Projects

A portable C++ library for digital signal processing and machine learning inferencing integrated as a native Zephyr RTOS module. It enables the deployment of optimized ML models across hundreds of hardware targets using the West build system and custom extension commands for automated model management.

TinyMLDelta provides a framework for incremental binary model updates on TinyML and embedded AI devices, reducing OTA bandwidth by up to 99% through diff patching. The project includes a Python-based patch generator and a platform-agnostic C runtime that manages atomic A/B model swaps, crash-safe journaling, and compatibility guardrails for TensorFlow Lite Micro models.

Noodle is a lightweight neural network inference engine designed to bring machine learning to resource-constrained microcontrollers like the Arduino Uno, Mega, and ESP32. By streaming activations and weights from external storage such as SD cards or flash memory, Noodle bypasses the typical RAM limitations that often prevent low-tier MCUs from running complex models.

A bare-metal application for the STM32L475VGT that performs real-time keyword recognition using TensorFlow Lite for Microcontrollers. It utilizes a quantized convolutional neural network to classify audio commands, supported by hardware-accelerated CMSIS libraries.

An Arduino core implementation for Silicon Labs microcontrollers, enabling support for Matter, BLE, and low-power wireless applications. It provides a comprehensive set of libraries and drivers for EFR32 and MGM24 series hardware, integrating Silicon Labs' Gecko SDK and FreeRTOS for advanced protocol stacks.

A comprehensive suite of Python scripts and services designed to facilitate the end-to-end integration of AI models onto STM32 microcontrollers and microprocessors. It supports PyTorch, TensorFlow, and ONNX frameworks, providing tools for training, quantization, benchmarking, and automatic application code generation for various computer vision and audio use cases.

An ESP32-CAM based assistive technology project designed for the deaf and mute community. It leverages AI to provide real-time speech-to-text conversion and American Sign Language (ASL) to speech translation, integrated into a wearable smart glasses form factor.

A modular framework for running TensorFlow Lite Micro models on ESP32 devices within the ESPHome ecosystem. It provides specialized components for digital and analog meter reading, image processing, and active learning data collection, optimized with ESP-NN for high-performance edge AI.

A specialized framework for training and deploying BirdNET audio classification models on the STM32N6570-DK development board. It provides a complete pipeline from dataset preparation and TensorFlow training to optimized NPU deployment using STMicroelectronics' X-CUBE-AI tools.

A modified Nintendo Wii Nunchuk controller powered by an ESP32 running TensorFlow Lite for Microcontrollers. It enables gesture-based commands via the joystick, supporting multiple action profiles like BLE HID media controls and administrative functions.

A comprehensive resource repository for the Microchip SAMA5D27 Cortex-A5 MPU, providing extensive documentation and guides for Linux development, bare-metal programming, and various RTOS ports. It features in-depth tutorials for NuttX, RT-Thread, and ThreadX, including driver integration for LVGL, EtherCAT, and TensorFlow Lite.

A port of the Tensorflow Lite Micro inference framework for the RT-Thread real-time operating system. It enables the deployment of deep learning models on resource-constrained embedded systems, featuring optimizations for ARM Cortex-M cores via CMSIS-NN and support for multiple hardware platforms including STM32 and Raspberry Pi.

A port of the TensorFlow Lite Magic Wand gesture recognition example to the mbed K66F platform. It utilizes mbed OS and TensorFlow Lite for Microcontrollers to perform real-time gesture detection on embedded hardware, covering the full pipeline from data collection to deployment.

A comprehensive collection of MicroPython scripts for the MaixPy platform, targeting Sipeed hardware based on the Kendryte K210 RISC-V SoC. It includes examples for machine vision, AI acceleration, hardware peripheral control, and multimedia processing.

A Bluetooth-enabled macro keyboard featuring gesture recognition powered by TensorFlow Lite for Microcontrollers on an ESP32. It combines capacitive touch sensors, physical buttons, and an OLED display to provide customizable computer control via BLE HID and AutoHotkey integration.

A comprehensive sample project for the NXP i.MX RT1060 Evaluation Kit (EVK) featuring a dual-stage architecture with a dedicated bootloader and a feature-rich application. It integrates Amazon FreeRTOS with a full middleware stack including lwIP networking, LVGL graphics, and FatFs storage for ARM Cortex-M7 development.

An example project demonstrating how to run the Fashion MNIST TFLite model on an ESP32 microcontroller using the ESP-IDF framework. It utilizes TensorFlow Lite Micro for edge AI inference, allowing the device to classify clothing items locally.

Anjay-zephyr-client is a collection of LwM2M client samples based on the Anjay Zephyr Module and Zephyr RTOS. It demonstrates how to implement LwM2M functionality, including sensor reporting, FOTA updates, and GPS tracking across various hardware platforms like nRF9160 and ESP32.

A curated collection of high-quality resources, libraries, tools, and learning materials for the Zephyr RTOS ecosystem. It serves as a comprehensive directory for developers looking for official documentation, community-supported libraries, and hardware platforms compatible with Zephyr.

SnoreSense is a TinyML application designed for the Seeed XIAO MG24 Sense to detect snoring in real-time. It utilizes TensorFlow Lite Micro for on-device inference, processing audio from the onboard PDM microphone to provide a privacy-focused, low-power sleep monitoring solution.