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
The accuracy of TSM-SELFY-R18/R50 on sth-sth v1 #1
Comments
To verify the effectiveness of the SELFY model, I also conduct the experiment on TSM-R18 and get the same results with SELFY-TSM counterpart. This SEFLY module does not seem to provide performance improvements. So I check your codes and comparison with the official TSM codebases. I find you changed the data augmentation (e.g., selective Flip, resize images to 240x320). I just simply add these two augmentations to the original TSM, then get an unbelievable strong TSM baseline (+2% on sth-sthv1). |
Hello whwu95, Thanks for your attention. |
Yes, I have utilized the 'use_corr_sampler=False', but I get only 51.467% on sthv1, which is 52.5% on the paper. |
Hi @whwu95, FileNotFoundError: [Errno 2] No such file or directory: '/home/something-something-v2/frames/177891/img_00003.jpg' Here's my log file : |
Hi, I have downloaded your official codes to reproduce the results of TSM-SELFY-R18 and R50 on the something-v1 dataset.
My top-1 accuracy of the R18 model is 42.6%, which is 48.4% in your paper.
And the top-1 accuracy of the R50 model is 48.3%, which is 52.5% in your paper.
Attached are my log files.
Did I get something wrong?
selfy-tsm50.log
selfy-tsm18.log
The text was updated successfully, but these errors were encountered: