Video Patterns
#4
by
Scofano
- opened
First, congratulations. It is the best local video model so far.
Is there a way to avoid (smooth?) those video patterns (like a grid over the video)?
These are artifacts, most likely from the perceptual loss used during training.
We are trying to reduce them with better training schemes.
From some attempts we made, you can use enhancers/upsamplers that are generative themselves to get rid of the artifact, but you pay in inconsistency.