Wang Xinyu

.cu contains CUDA code, not a pure cpp code.

@TengjiaoYuanyangguo This ROS integration is not related to tensorrtx repo. You need to fix the issue by yourself.

INT8 has lower accuracy, this is normal, maybe you can try other calib data or more calid data, or try INT8 QAT method. The inference speed depends on your device,...

Did you try the latest gen_wts.py?

Our code doesn't support windows directly, some changes are required.

Thanks for the PR, but the v4.0 is a tag, not branch, so we won't update it anymore. We only make changes to the master branch.

Which pytorch trained model were you using?

@dubeamit Maybe your pytorch version it too new, at that repo, they were using 1.1. Can you try 1.1 or 1.3?

Since it worked in one of your environment, then it's not the code issue. You need to check the gpu driver, cuda, tensorrt installation, etc.

Looks like your pytorch model doesn't match with the Trt model.