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

Warning:Offset mean is XXX,larger than 100 #30

Closed
porcofly opened this issue Aug 3, 2020 · 2 comments
Closed

Warning:Offset mean is XXX,larger than 100 #30

porcofly opened this issue Aug 3, 2020 · 2 comments

Comments

@porcofly
Copy link

porcofly commented Aug 3, 2020

2020-08-03 15-37-21屏幕截图
2020-08-02 17-37-36屏幕截图

Thank you for open source your code, when train the model on Vimeo 90k,this warning occurred.
It first appeared after the first decay of the learning rate, and then the offset mean will become larger and larger. Have you ever encountered this situation? Is this normal in your training?

@Mukosame
Copy link
Owner

Mukosame commented Aug 6, 2020

Hi @porcofly , basically this warning will not kill your training process, but you have to be careful of it --- since the Vimeo90k is only ~480 in size, having a 200 offset probably means your model doesn't converge.
I also encountered such situations before. In fact, such issues also happen to other works that use DCN. You can try some tricks like warm-up, or initialize with our pretrained weights to provide a safer start point for you model to prevent it from happening again.
Hope these can solve your questions!

@porcofly
Copy link
Author

Hi @porcofly , basically this warning will not kill your training process, but you have to be careful of it --- since the Vimeo90k is only ~480 in size, having a 200 offset probably means your model doesn't converge.
I also encountered such situations before. In fact, such issues also happen to other works that use DCN. You can try some tricks like warm-up, or initialize with our pretrained weights to provide a safer start point for you model to prevent it from happening again.
Hope these can solve your questions!

Thanks for your reply, I changed the learning rate to 1 e-4 and train from scratch.Then this problem seem be solved

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