search

eqs / DeepLabCut-Docker

Dockerfile for DeepLabCut 2.0 (https://github.com/AlexEMG/DeepLabCut)
8 stars 3 forks source link

All GPUs are used when analyze_videos #1

Open DonnieKim411 opened 5 years ago

DonnieKim411 commented 5 years ago

Hi eqs,

I know you are not directly involved in DLC development, but I found some weird behavior that is described here: https://github.com/AlexEMG/DeepLabCut/issues/233

Have you experienced such behavior on your machine? I am using your dockerfile.

Cheers,

eqs commented 5 years ago

Hi @DonnieKim411,

Sorry for the late response. Thank you for using my Dockerfile.

Have you experienced such behavior on your machine? I am using your dockerfile.

Yes, the weird behavior that both GPUs occupy all memory appeared in our environment.

When the weird behavior appeared, I addressed to it by adding the following code to the top of my .ipynb file. This code controls GPU visibility in the notebook.

import os
os.environ['CUDA_VISIBLE_DEVICES'] = '0' # change the number to you want to use

In our case, we parallelized multi-tasks (e.g. training model and analyzing videos) by using the environment setting.

If you are still interested in this issue, please try above.

MMathisLab commented 4 years ago

Hey, I know this is old, but this was fixed in a more recent DLC version, i.e. beyond 2.0.9 I believe, so I think this issue can be closed.