-
Notifications
You must be signed in to change notification settings - Fork 97
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Computation time for i3d features #120
Comments
Hi, sorry to hear about that. Can you give me a bit more information? Did you try to divide into 4 subsets (1 per GPU, instead of running 4? on one GPU)? It is faster with Do you see your GPU being 100% utilized? |
Yes. I also divided into 4 subsets running on each of the gpus along 4 more subsets on each gpu. And with this split gpu utilization was 100%. I need both rgb and flow feature so did not try separately. Do you recommend this? |
One more issue is that out of 16 parallel jobs submitted, it has computed features for first 4 jobs running one of each of 4 GPUs. Reaming 12 jobs, which were computing features, did not save the feature even though job was running full time. All jobs were writing the output features to same folder. It means if there is single GPU, I can't run multiple jobs to process more number of videos? Am I right? Can you suggest a solution as it is really taking lots of time. |
Try using one process per GPU. Did it improve speed? |
Yes. it improves but still it is quite slow for untrimmed videos. |
Hi. Thanks for awesome work.
I am not able to extract visual features in an efficient way. Its taking too much time even on GPUs. I am extracting visual features on 278 videos of 4-5 minutes duration by dividing into 16 parallel subsets simultaneously on 4GPUs of 24GB RAM. It has extracted features for only 60 videos in 24 hours. Can you suggest an efficient way for the same?
Thank you.
The text was updated successfully, but these errors were encountered: