fps really low

[maylad31]

Hi i installed gpu version following the instructions but I get very slow fps.It seems my gpu is not being used.Please help.

[ohadlahav]

Hi, i would assume that one will need to use nvidia-docker to fully leverage GPU in docker environment.
I did manage to install and use successfully yolo34py-gpu on a nvidia/cuda/cudnn local environment.

[maylad31]

hi could you please give me more details about fps in your local environment. And did you install using pip or build from source?

[pherris]

For me it’s more like spf (about 6) - running on a Mac.

…

On Sun, Dec 2, 2018 at 10:17 AM Mayank Laddha ***@***.***> wrote: hi could you please give me more details about fps in your local environment. — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#53 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AB_O9vBxmB8O13QmFcCkOovGH4PjqBKUks5u0_0fgaJpZM4Y4E42> .

[madhawav]

@maylad31 For the GPU version to work in a stand-alone machine, you will need CUDA SDK installed. Also, you will need a Nvidia Cuda enabled GPU. If you are running on docker, nvidia-docker would be required.

Do you have a Cuda enabled GPU?
Are you running in Docker?
Do you have CUDA installed?

[maylad31]

thanks sorry for the late reply you may close this issue yes i solved the problem i had cuda installed and path and everything was set but still it wasn't working so i updated my nvidia driver reinstalled cuda and it worked thanks i would like to ask why the darknet fps is more than python wrapper i understand it would be slow in python but with darknet i get 18 fps and with python wrapper it drops to around 11-12 fps

[madhawav]

Hi @maylad31,
There is a drop in fps caused by conversion of python images (numpy arrays) to darknet images. The drop depends on the resolution of images.