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

Wrong MoCo-v3 pretrained weights? #53

Open
tsly123 opened this issue Oct 27, 2023 · 0 comments
Open

Wrong MoCo-v3 pretrained weights? #53

tsly123 opened this issue Oct 27, 2023 · 0 comments

Comments

@tsly123
Copy link

tsly123 commented Oct 27, 2023

Hi,

Thank you for the greate work.
This repo (VPT) provides the link to download the MoCo-v3 pretrained weights, named as mocov3_linear-vit-b-300ep.pth.tar. According to the MoCo-v3 github page (at the last table), it's the linear tuning ckpt not the SSL pretrained which is named as vit-b-300ep.pth.tar.

If I understand correctly, the Moco-based results in the VPT paper are got from fine-tuning experiments on the linear classification ckpt not the SSL pretrained one. So the question is: Have you tried to experiment with SSL ckpt? and Is there any different between finetuning with the SSL ckpt and linear classification ckpt?
Or am I missing something here?

Thank you

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