TensorRT conversion

[yangjin07]

@maudzung Hi，Maudzung, nice work! I'm trying to use TensorRT to speed up the inference. Do you have some scripts that transfer the Complex-YOLOv4-Pytorch from pytorch(pth file) to TensorRT?

Thanks!

[maudzung]

Hi @yangjin07 I didn't do it. Please contribute to the repo if you made it. Thank you so much!

[PengfeiMa]

Hi, @maudzung I have tried to use the TRT to speed up the inference, but there is some trouble in the 'yolo_layer'. When I export ONNX model with yolo_layer, the 'x, y, w, l, im, re' in ONNX inference outputs are 0. When I export pytorch model to ONNX model, there are just some warnings, no errors. So I export the ONNX model without the 'yolo_layer', and add the yolo_layer in the post_process. That's work! This is my ONNX model without yolo_layer.

[maudzung]

Hi @PengfeiMa That sounds good. Can you make a pull request to the repo? I'll check and merge your work into the repo. Thank you so much!

[YinRui1991]

@PengfeiMa Could you share some ideas about how to export onnx model without 'yolo layer' ? Thanks a lot.

[MaxZJF]

Hi @PengfeiMa Can you share how to convert PTH to onnx without‘yolo layer’ ? Thank you so much!