Is it possiable to use pixart as base model to train a image super resolution model? #100
rainsoulsrx
started this conversation in
General
Replies: 0 comments
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Uh oh!
There was an error while loading. Please reload this page.
-
Hi, just like this work, https://github.com/Fanghua-Yu/SUPIR, they use SDXL as base model. But SDXL has some limations, such as it can only take into 77 tokens, and its vae is not friendly for small objects as words and small human faces. So is it possiable to use pixart model as the base model for image super resolution work?
Beta Was this translation helpful? Give feedback.
All reactions