Ultra-lightweight human detection for XIAO

Summary

This is a sample running PINTO0309's super cool lightweight person detection model “UHD” on the Seeed Studio XIAO ESP32S3 Sense. This model can detect people from small 64×64 pixel images and is extremely fast.

20260212A.mp4

Prerequires

• Hardware

  • x1 Seeed Studio XIAO ESP32-S3 Sense
  • x1 Round Display for Seeed Studio XIAO

• Software

  • ESP-IDF v5.5

Setup

1. Clone this repository.

2. Copy the model file.

   Download and extract the model file.

   • ultratinyod_anc8_w32_64x64_opencv_inter_nearest_static_nopost.tar.gz

   \MODELS
   └─uhd
       └─ultratinyod_anc8_w32_64x64_opencv_inter_nearest_static_nopost
               ultratinyod_anc8_w32_64x64_opencv_inter_nearest_static_nopost.espdl
               ultratinyod_anc8_w32_64x64_opencv_inter_nearest_static_nopost.info
               ultratinyod_anc8_w32_64x64_opencv_inter_nearest_static_nopost.json
               ultratinyod_anc8_w32_64x64_opencv_inter_nearest_static_nopost.onnx
               ultratinyod_anc8_w32_64x64_opencv_inter_nearest_static_nopost_anchors.npy
               ultratinyod_anc8_w32_64x64_opencv_inter_nearest_static_nopost_wh_scale.npy
   

3. Set the target to esp32s3.

4. Build, flash and monitor.

License

• espressif/esp32-camera (Apache-2.0)
• lovyan03/LovyanGFX (FreeBSD)
• espressif/esp-dl (MIT)
• PINTO0309/UHD (MIT)
• PINTO0309/esp-who (folk) (MIT)
• SeeedJP/UHD-for-XIAO (MIT)