tfjs_Edge-computing-by-inference-AImodel-on-WebGL

Test tensorflow js for custom pretrained model(Pytorch U-2NET)

All valuable implements are in test.js

I recommend you to notice these references.

reference

• Tensorflow.js
• Tfjs api
• pytorch to onnx
• onnx to tensorflow
• tensorflow to tfjs
• tensorflow lite quantization

What I did

• Convert pretrained Pytorch model to onnx
• Convert onnx model to tensorflow(saved model)
• Convert tensorflow model to tfjs, tflite
• Load tfjs model by Tensorflow.js Graphmodel
• Using WebGL backend, implement inference code
• Load tflite model by tflite-alpha module, but not implemented yet(see the issues below)
• Quantization Tfjs and Tflite

TODO

I found these issues, at now tfjs doesnt surpport gpu implementation for tensorflow lite. So I need update so that I can use tflite models.(tensorflow js model can used)