Model deployment regarding flashocc #101
Description
Thank you for your open-source work. The link you provided previously for the TensorRT (C++) example is now invalid.
I tried to run the Demo using the following link, and repeated the experiments in multiple environments.
https://github.com/hituwangli/TRT_Flashocc
The code, data, and ONNX model remained consistent. However, there is a significant accuracy difference between PyTorch and TensorRT, and they are almost unmatched.
Could you please reactivate that link?
https://github.com/drilistbox/TRT_FlashOcc
Regarding the model deployment environment, such as CUDA, cuDNN, and TensorRT versions, do you have any recommended configurations?
I am looking forward to your reply. Thank you again.