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

How the identity is kept as much as possible? #53

Closed
JcccKing opened this issue Nov 17, 2022 · 2 comments
Closed

How the identity is kept as much as possible? #53

JcccKing opened this issue Nov 17, 2022 · 2 comments

Comments

@JcccKing
Copy link

Hello Researcher Yang.
Thank you for your excellent work!!! I followed your training pretty well and kept my style great! But the identity and structure change a lot, do you have any better solution for the difference in identity due to PSP inverse mapping? Or which part in between you need to fine tune according to the dataset, please give me some hints, thanks!
image

@williamyang1991
Copy link
Owner

  • You can use a lower --weight. For example, --weight 0 0 0 0 0.5 0.5 0.75 1 1 1 1 1 1 1 1 1 1 1 will only transfer colors.
  • You can also perform optimization over the W+ to make better reconstruction.
  • Based on your pretrained DualStyleGAN, you can also try VToonify

@JcccKing
Copy link
Author

Hello Researcher Yang.
Thank you very much for your prompt reply, I will try the above method and follow up by trying video style transfer. Thank you for your excellent work !!!

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