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

Reproducability #2

Open
BenediktAlkin opened this issue Feb 4, 2024 · 0 comments
Open

Reproducability #2

BenediktAlkin opened this issue Feb 4, 2024 · 0 comments

Comments

@BenediktAlkin
Copy link

BenediktAlkin commented Feb 4, 2024

Hi,

I really like the idea of your paper to use a GAN discriminator as feature extractor for perceptual losses to improve MAE pre-training.
I tried to play around a bit with the idea myself but the codebase is unfortunately incomplete and the implementation details in the paper are lacking to say the least.

Are there any plans to update the repo and/or publish the trained models?

In my opinion, your strong claims in the paper require at least some form of reproducability. Claiming insanely good results without anything to back it up is quite questionable. To clarify: your paper claims a ImageNet-1K finetuning accuracy of 88.1% with a ViT-L/16, which would be 0.3% better than the ViT-H/14_448 trained from the original MAE paper.

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