Slow onnxruntime inference speed with u7 segmentation

[laitathei]

I follow https://github.com/WongKinYiu/yolov7/issues/771 to change the code I have tried to export pytorch to onnx format and I want to test with some video for inference. I found that the inference speed reach around 465.7ms. I think that the onnxruntime should not have this performance. Does anyone know what how to fix it?

[sadimoodi]

I am also experiencing slow performance of yolov7 when converting to onnx, had to revert back to pytorch format

[NgJoshuaJeremiah]

Same here, I'm facing a 100% drop in inference speed, from 10ms to 20ms when I use the converted onnx yolov7 model.