search

zsyOAOA / ResShift

ResShift: Efficient Diffusion Model for Image Super-resolution by Residual Shifting (NeurIPS@2023 Spotlight, TPAMI@2024)
Other
870 stars 47 forks source link

ResShift without the autoencoder #52

Closed chandrasekaraditya closed 5 months ago

chandrasekaraditya commented 6 months ago

Hi

Thanks for the great work! I have a small query on training the model without the autoencoder. If i directly declare it to be none in the config file, i.e.,

autoencoder: None

it errors out due to the other dependencies on the autoencoder config params in the script. I also tried only making the target none and leaving the rest, but doesn't seem to work. Could you please guide me as to how to train ResShift without the autoencoder, basically on the image space instead of the latent space? Thanks a ton, and I hope to hear from you soon.

chandrasekaraditya commented 5 months ago

@zsyOAOA, any advice would be helpful, Thanks!

zsyOAOA commented 5 months ago

You need to rewrite part of the code involving the autoencoder. I have ever attempted to conduct experiments on the pixel space. Even it works well, but costs too much GPU memory. The related code of mine is lost, thus I can't share the code with you.