BiSeNet icon indicating copy to clipboard operation
BiSeNet copied to clipboard

Published 20 hours ago • verified publisher icon CoinCheung

internal error

Open Datou98 opened this issue 2 years ago • 11 comments

2022-09-08 15-05-22 的屏幕截图 Hello, I would like to ask how to solve this problem, about Tensorrt versions ?

Datou98 avatar Sep 08 '22 07:09 Datou98

What is the configuration of your platform? Did you modify the code? Please align the cuda/cudnn/tensorrt/pytorch/cmake version as mine listed in readme.

CoinCheung avatar Sep 08 '22 09:09 CoinCheung

What is the configuration of your platform? Did you modify the code? Please align the cuda/cudnn/tensorrt/pytorch/cmake version as mine listed in readme.

ubuntu 18.04  GeForce RTX 3060 Driver Version 470.94 cuda 11.1 cudnn 8.0.5 tensorrt 8.2.5.1 cmake 3.16.2 opencv 3.4.15 I did not modify the code

Datou98 avatar Sep 08 '22 10:09 Datou98

Did you export onnx model with the same command as in README?

CoinCheung avatar Sep 08 '22 11:09 CoinCheung

Did you export onnx model with the same command as in README?

yes，I have ported onnx mode,Now I'm upgrading cuda, try again

Datou98 avatar Sep 08 '22 11:09 Datou98

I mean, did you add --aux-mode eval when you export onnx model?

CoinCheung avatar Sep 08 '22 11:09 CoinCheung

I mean, did you add --aux-mode eval when you export onnx model?

yes,I add

Datou98 avatar Sep 08 '22 11:09 Datou98

I mean, did you add --aux-mode eval when you export onnx model? Also, I would like to ask do I need to install onnx?

Datou98 avatar Sep 08 '22 11:09 Datou98

I tried again on my platform, and it works as expected. Maybe you should align your platform configuration as mine, especially the cuda/cudnn version.

CoinCheung avatar Sep 08 '22 11:09 CoinCheung

I tried again on my platform, and it works as expected. Maybe you should align your platform configuration as mine, especially the cuda/cudnn version.

What are the python and pytorch versions you are using？

Datou98 avatar Sep 08 '22 11:09 Datou98

I use pytorch3.8 from miniconda and pytorch1.11.0(cuda11).

CoinCheung avatar Sep 08 '22 12:09 CoinCheung

I use pytorch3.8 from miniconda and pytorch1.11.0(cuda11). 2022-09-08 22-16-45 的屏幕截图

Thanks, I can convert model_final_v2_city.pth to trt by upgrading cuda from 11.1 to 11.3 and cudnn from 8.0.8 to 8.2.1, which is normal for FPS to 118.283 in an environment where the CPU is R7-5800H and the GPU is 3060?

Datou98 avatar Sep 08 '22 14:09 Datou98