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

About reproduction #6

Open
sHaol0w opened this issue Oct 30, 2023 · 3 comments
Open

About reproduction #6

sHaol0w opened this issue Oct 30, 2023 · 3 comments

Comments

@sHaol0w
Copy link

sHaol0w commented Oct 30, 2023

Hi, I have tried to reproduce your excellent work LiveHand, but I encountered some problems in the rendered border. Do you have any ideas?

livehand1
1698662661956

@amundra15
Copy link
Owner

Hi,

Are these renderings from the model weights we provide or did you train a new model?

@sHaol0w
Copy link
Author

sHaol0w commented Oct 31, 2023

Hi,

Are these renderings from the model weights we provide or did you train a new model?

I trained a new model and used my data preprocessing, which is fine for my method and other method.
If you do not see this ever, then there may be some problems in my side?

@amundra15
Copy link
Owner

amundra15 commented Oct 31, 2023

We never saw these artifacts in the final training configuration. In one of the older configurations where we tried an additional BG loss resulted in somewhat similar artifacts.

I assume that the results you uploaded are for the super-resolved image. In that case, I would also save the low-res image from the neural rendering part to see if it is present there as well. Otherwise, the issue comes from an improperly trained SR module.

Another source of error could be incorrect supervision. I think I binarize the loaded BG mask and multiply it with the RGB image to get the GT image. If there is some issue here, that can cause this error too.

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