av-savchenko/face-emotion-recognition

The issue with converting .pt to .onnx.

Closed this issue · 4 comments

xx674967 commented

Hi, excellent work, and thank you for sharing!
I have a question regarding converting .pt to .onnx using your convert_pt_to_onnx.py script. I encountered an issue where, despite installing timm==0.4.5, it gives an error stating no timm.layer. I understand this is likely due to the timm version, but updating to a newer version causes other unexpected issues. Do you have any suggestion? thanks

av-savchenko commented

Try timm 0.9.7 and let me back full stack trace if this does not work for you

xx674967 commented

Try timm 0.9.7 and let me back full stack trace if this does not work for you

Thank you for your quickly response. After I changed the timm version to 0.9.7, I encountered this error:
RuntimeError: Input type (torch.FloatTensor) and weight type (torch.cuda.FloatTensor) should be the same or input should be a MKLDNN tensor and weight is a dense tensor.
I've tried modifying the input tensor to .cuda(), but the problem persists. Could you please help me find a solution? Thank you very much.

xx674967 commented

I found a fascinating issue: converting to ONNX works fine on Windows, but on Linux, various strange errors occur. I'm not sure of the exact reason. Anyway, I will close this issue.

ceyxasm commented

Worked with timm==0.9.2.
Executed flawlessly.