
I trained a model to cleanup the pseudo pixel arts:
Left is 128×128 resampled generated images.
Middle is the cleaned up version on top of white bg.
Right is the cleaned up version on top of black bg.
From zoomed out view, I think these characters look decent. However, there are quite a bit of alpha errors.
- We need more various data
- See if we can use pretrained model
Comments
Join the discussionNo comments yet. Be the first to comment.



