What problem does it solve? Developing custom object detection for Espressif chips requires coordinating two separate toolchains—Python training/quantization on PC and ESP-IDF C++ firmware on device—with many failure-prone details like custom module registration, ONNX opset constraints, calibration data, and Kconfig model placement. This Skill guides AI agents through the entire esp-detection workflow using only verified repository source code. ## Core Features & Use Cases - Scenario Recipes: Nine step-by-step recipes covering environment setup, YOLO dataset preparation, square and rectangular (rect=True) training, ONNX export (opset 13), INT8 quantization to ESP-DL .espdl via esp-ppq, quantized mAP evaluation, one-command pipeline, and firmware deployment. - API & Config Reference: Complete signatures for Train/Export/quant_espdet, custom network modules (DSConv, ESPBlock, ESPDetect), C++ classes (ESPDetDetect), plus Kconfig, sdkconfig, and partition table references. - Pitfall Prevention: 26 documented pitfalls with WRONG/CORRECT code comparisons, such as injecting custom_parse_model before training and enabling RGB565 big-endian on ESP32-S3. - Use Case: Train a custom cat detector at 224x224, quantize it to INT8, generate an ESP-IDF project automatically via espdet_run.py, and flash it to an ESP32-P4 board to see detection results over serial. ## Quick Start Ask the AI agent to train and deploy a custom object detection model on ESP32-P4 using esp-detection, and it will follow the all-in-one pipeline recipe from dataset preparation through firmware flashing.
