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

s and m in ArcFace loss #8

Closed
wjtan99 opened this issue Oct 23, 2023 · 4 comments
Closed

s and m in ArcFace loss #8

wjtan99 opened this issue Oct 23, 2023 · 4 comments

Comments

@wjtan99
Copy link

wjtan99 commented Oct 23, 2023

Thanks for sharing your great work.
I noticed that you have s=45, m=0.1 by default in the loss function. In your configx.py, you have s=45, m=0.3,0.4 etc. In the original ArcFace paper, the default s=60, m=0.5. Did you optimize the s and m in your work?

@ChristofHenkel
Copy link
Collaborator

yes, s and m are important hyperparameter, which we optimized

@wjtan99
Copy link
Author

wjtan99 commented Oct 23, 2023

What is the best value of your optimization? You have quite a few configx.py, so I am not sure which one is your final choice.

@ChristofHenkel
Copy link
Collaborator

If I remember correctly, we used slightly different m depending on if we also trained on GLDv1. Having slightly different m values per config, also helped diversity in our ensemble.

@wjtan99
Copy link
Author

wjtan99 commented Oct 23, 2023

Thanks a lot for your prompt response.

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

2 participants