search

zhang-zx / SINE

This respository contains the code for the CVPR 2023 paper SINE: SINgle Image Editing with Text-to-Image Diffusion Models.
https://zhang-zx.github.io/SINE/
MIT License
181 stars 11 forks source link

About Patch-based Finetune #11

Open wty-ustc opened 1 year ago

wty-ustc commented 1 year ago

Thanks for your very great work! In the paper, you use patch-based finetune to make the model support large resolution editing, but the original pre-trained model is not finetuned. My concern is that when performing model guidance will the blending of the two predicted noises create artifacts due to mismatch?

Ed-ivan commented 1 year ago

I think that is a good question . And I feel a little confused that why the cropped image is resized to [512,512]