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.

Sign up for GitHub

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

Jump to bottom

When the number of classes is very large, the loss function does not converge #5

Open
zzzmm1 opened this issue Feb 8, 2023 · 1 comment
Open

When the number of classes is very large, the loss function does not converge #5

zzzmm1 opened this issue Feb 8, 2023 · 1 comment

Comments

@zzzmm1
Copy link

zzzmm1 commented Feb 8, 2023

In face recognition, when the number of classes exceeds 10 million and K=128, the CosFace/ArcFace loss function hardly converges, due to the fixed binary orthogonal codes.
Any suggestions to help with training?
Thanks a lot!
@kamwoh
Copy link
Owner

kamwoh commented Mar 4, 2023

I think you can try adding some non-linear layers before the hash layer. sometimes it overfits faster so I think it might help to converge.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Assignees
No one assigned
Labels
None yet
Projects
None yet
Milestone
No milestone
Development

No branches or pull requests
2 participants
@kamwoh @zzzmm1