BlindDiff

Introduction

This is the official code of our work BlindDiff: Empowering Degradation Modelling in Diffusion Models for Blind Image Super-Resolution.

The pretrained models are Baidu Disk and Google Drive

This repo is built on the basis of BasicSR and guided-diffusion, thanks for their open-sourcing!

Environment

• Python3
• pytorch>=1.7

  Installations

  Run the command:

  pip install -r requirement.txt

  and

  python setup.py develop

  Train

  1. Download trainning dataset DIV2K and Flickr2K for the natural images and FFHQ for the face images.
  2. Configure options/train.yml for your training.
  3. Run the command:

     python basicsr/train.py -opt=options/train_setting.yml

     Test

  4. Configure options/test.yml for your training. The testing dataset used in the paper is here.
  5. Run the command:

     python basicsr/train.py -opt=options/test.yml

     Citation

     If you find our work useful in your research or publications, please consider citing:

     @article{li2024blinddiff,
     title={BlindDiff: Empowering Degradation Modelling in Diffusion Models for Blind Image Super-Resolution},
     author={Li, Feng and Wu, Yixuan and Liang, Zichao and Cong, Runmin and Bai, Huihui, Zhao, Yao and Wang, Meng},
     journal={arXiv preprint arXiv:2403.10211},
     year={2024}
     }