Skip to content
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

batch_size doesn't affect evaluation #225

Open
SuroshAhmadZobair opened this issue Jan 14, 2024 · 0 comments
Open

batch_size doesn't affect evaluation #225

SuroshAhmadZobair opened this issue Jan 14, 2024 · 0 comments

Comments

@SuroshAhmadZobair
Copy link

Hi

Thanks for your work on this repo.
I am trying to do evaluation on a few models I trained in the past but it takes 3 Hrs for evaluation on each model. I am using RTX 3090 with batch size 128. yes, I have tried changing the batch size it seems like its not affecting the evaluation time. 14GBs of my GPU is also free. No matter wat batch size or num_workers i use, I still get the following timing for evaluation.

[01/14 14:27:12 d2.evaluation.evaluator]: Inference done 11/23852. Dataloading: 0.0006 s/iter. Inference: 0.0625 s/iter. Eval: 0.3715 s/iter. Total: 0.4347 s/iter. ETA=2:52:42

[01/14 14:27:18 d2.evaluation.evaluator]: Inference done 24/23852. Dataloading: 0.0007 s/iter. Inference: 0.0572 s/iter. Eval: 0.3645 s/iter. Total: 0.4225 s/iter. ETA=2:47:46

[01/14 14:27:23 d2.evaluation.evaluator]: Inference done 36/23852. Dataloading: 0.0007 s/iter. Inference: 0.0566 s/iter. Eval: 0.3713 s/iter. Total: 0.4287 s/iter. ETA=2:50:10

How can i speed up this process?
Here is the cmd i use:

python train_net.py --config-file configs//instance-segmentation/maskformer2_R101.yaml --eval-only SOLVER.IMS_PER_BATCH 128 MODEL.WEIGHTS for_eval_testing/model_0154999.pth

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Labels
None yet
Projects
None yet
Development

No branches or pull requests

1 participant