README.md

yolov9-wholebody25-tensorflowjs-web-test

A test environment running yolov9-wholebody25 on TensorFlow.js.

https://github.com/PINTO0309/PINTO_model_zoo/tree/main/459_YOLOv9-Wholebody25

• Model conversion

  pip install -U onnx2tf
  
  pip install -U --no-deps \
  tensorflowjs \
  tensorflow_decision_forests \
  ydf \
  tensorflow_hub
  
  onnx2tf -i yolov9_n_wholebody25_post_0100_1x3x480x640.onnx -cotof -dgc
  
  tensorflowjs_converter \
  --input_format tf_saved_model \
  --output_format tfjs_graph_model \
  saved_model \
  tfjs_model
  

  

• Starting the inference web server

  python -m http.server
  

• Displaying a web page (with TensorFlow.js + WebGL backend (AUTO) + USB Camera)

  If your browser cache is enabled, model loading and inference may fail. If this does not work, try starting your browser in an incognito window or in secret mode before accessing the URL below.

  http://localhost:8000/test-tfjs-usbcam.html

  or

  http://localhost:8000/test-tfjs-usbcam-low.html

• Results

  • Normal Resolution Mode 640x480

    Kazam_screencast_00012_.mp4

  • Low Resolution Mode 320x256

    Kazam_screencast_00014_low.mp4

• Check if your browser supports WebGL hardware acceleration

  https://get.webgl.org/

  

• Extra

  This is a game where you interactively draw a polygonal area with a mouse click or double click and a message is displayed when your hand enters the area.

  http://localhost:8000/test-tfjs-usbcam-inner-polygon.html

  Kazam_screencast_00021_.mp4