onnx-tensorflow onnx to pb parse conv(padding=='same') as 2 seperate layer (pad + conv)

onnx to pb parse conv(padding=='same') as 2 seperate layer (pad + conv)

Open huangteng opened this issue 4 years ago • 3 comments

Describe the bug

The tools parses conv(padding=='same') as 2 seperate layer (pad + conv) when converting from onnx to pb, this will affect the performance.

Python, ONNX, ONNX-TF, Tensorflow version onnx-tf: v1.7.0-tf-1.15 python: 3.6 onnx: 1.8.0 tensorflow: 1.15.0

Jul 13 '21 03:07 huangteng

Thanks for bringing up the performance concerns. We are short in resources, so contributions are welcome!

Aug 02 '21 22:08 chinhuang007

Thanks for bringing up the performance concerns. We are short in resources, so contributions are welcome!

Indeed I have a fix for this :) So I will try to submit a push request. Do you need to set my account first ?

Aug 12 '21 13:08 huangteng

Yes, please set up a github account. Thanks in advance!

Aug 12 '21 15:08 chinhuang007