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

Difference between the code and the paper about the training #37

Open
weihaosky opened this issue Dec 10, 2024 · 0 comments
Open

Difference between the code and the paper about the training #37

weihaosky opened this issue Dec 10, 2024 · 0 comments

Comments

@weihaosky
Copy link

Thanks for this excellent work!

  1. In the paper, it writes 50% images from FFHQ and CelebA, 40% images from MEAD, 10% images from LRS3. However, in the code the percentage for FFHQ, CelebA, MEAD, LRS3 are 30%, 30%, 20%, 20%, respectively. May I ask why is this difference? which one should I follow?
  2. In the paper, it writes the pretraining takes 60,000 iterations while the core training takes 250,000 iterations. However, in the code, the config for pretraining employs 300 epochs, which corresponds to around 468,000 iterations, and the config for core training employs 50 epochs, which corresponds to around 78,000 iterations. May I ask why is this difference? which one should I follow?

Many thanks!

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