Generation of `saved_model` with `GroupConvolution` and Full Integer Quantization without `--dgc`

PINTO0309 / onnx2tf

Self-Created Tools to convert ONNX files (NCHW) to TensorFlow/TFLite/Keras format (NHWC). The purpose of this tool is to solve the massive Transpose extrapolation problem in onnx-tensorflow (onnx-tf). I don't need a Star, but give me a pull request.

MIT License

706 stars 73 forks source link

Generation of `saved_model` with `GroupConvolution` and Full Integer Quantization without `--dgc` #655

Closed PINTO0309 closed 4 months ago

PINTO0309 commented 4 months ago

1. Content and background

Generation of saved_model with GroupConvolution and Full Integer Quantization without -dgc, -osd.
Improved stability of Full Integer Quantization.
YOLOv9 with GroupConvolution + Dynamic Input [1, None, None, 3]

PINTO0309 / onnx2tf

Generation of `saved_model` with `GroupConvolution` and Full Integer Quantization without `--dgc` #655

1. Content and background

2. Summary of corrections

3. Before/After (If there is an operating log that can be used as a reference)

4. Issue number (only if there is a related issue)