Workaround to the problem of TensorFlow corrupting the FlatBuffer input/output order during INT8 quantization. - Githubissues

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

708 stars 73 forks source link

Workaround to the problem of TensorFlow corrupting the FlatBuffer input/output order during INT8 quantization. #651

Closed PINTO0309 closed 5 months ago

PINTO0309 commented 5 months ago

1. Content and background

Workaround to the problem of TensorFlow corrupting the FlatBuffer input/output order during INT8 quantization.
Determination of the number of inputs/outputs of the same shape.
Correct name discrepancies based on shape if multiple inputs/outputs shapes do not overlap.
However, if there are inputs/outputs containing undefined dimensions, workaround is skipped because correction is not possible.

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)

Input and Output Name Order Swapping with -coion option #650