search

zsyOAOA / ResShift

ResShift: Efficient Diffusion Model for Image Super-resolution by Residual Shifting (NeurIPS@2023 Spotlight, TPAMI@2024)
Other
897 stars 49 forks source link

image size #85

Open MinjuJangg opened 3 months ago

MinjuJangg commented 3 months ago

I want to super-resolution the image from 512 to 1024, Should I train vqgan separately? or can i use exist vqgan?

zsyOAOA commented 3 months ago

We haven't trained a model for the x2 SR task. Sorry.

If you want to train such a x2 model, you can still use the existing vqgan but re-train the diffusion model.

MinjuJangg commented 2 months ago

Thank you for your reply. I tried it. It's trained well at size 256. But 1024 size is too big and it causes CUDA out of memory.

Is there any way to train High resolution image?

I saw another your comment that the model can super resolution for any resolution. But, it didn't worked well at not trained size.

zsyOAOA commented 2 months ago

For large resolution, we will first chop it into overlapped patches and process each patch separately. @MinjuJangg