Issue running denseflow with docker

[RabiaCodes]

trafficstars

I am trying to run denseflow on docker. But i get this error. How can i resolve this?

Input: docker run [docker_image:mytag] path/to/video -b=20 -a=tvl1 -s=1 -v Error: OpenCV(4.5.2) /opencv/modules/core/src/cuda_info.cpp:73: error: (-217:Gpu API call) CUDA driver version is insufficient for CUDA runtime version in function 'setDevice'

[elicz]

I have the same error too.

[RabiaCodes]

I couldnt resolve it on docker. So i set up denseflow on linux operating system.

On Wed, Apr 27, 2022, 4:37 PM elicz @.***> wrote:

I have the same error too.

— Reply to this email directly, view it on GitHub https://github.com/open-mmlab/denseflow/issues/62#issuecomment-1110653075, or unsubscribe https://github.com/notifications/unsubscribe-auth/AQCDSSPNA5AMK63ESEQXROTVHDVEHANCNFSM5RS7K56Q . You are receiving this because you authored the thread.Message ID: @.***>

[shink]

Hi, @RabiaCodes

I guess you should add the --gpus all parameter when running denseflow on docker.

sudo docker run --gpus all

But I didn't test it. I failed to build it locally. And has anyone successfully built it and pushed it to Docker Hub? Thank you!

[Permafacture]

In the Dockerfile, you have to set the CUDA and CUDNN versions such that it pulls a base container compatible with your graphics card, and you have to set the CUDA_GENERATION to be that of your card also. I have and RTX 3080 and so these were 11.4.0, 8, and Ampere respectively.

I was already using the --gpus all and I think I had to install some nvidia runtime thing for docker as well before I found that to not solve this and updated these values. I do think the nvidia runtime is necessary also but can't say for sure.