Home

Awesome

Coarse-to-Fine Mechanisms Mitigate Diffusion Limitations on Image Restoration (C2F-DFT)

Liyan Wang, Qinyu Yang, Cong Wang, Wei Wang, and Zhixun Su*

[2024-08-11] Our paper is accepted to Computer Vision and Image Understanding (CVIU).

Coarse Training Pipeline of Diffusion Transformer Model (DFT)

<img src = "https://github.com/wlydlut/C2F-DFT/blob/main/Figs/coarse.png#pic_center">

Fine Training Pipeline and Sampling Phase

<img src = "https://github.com/wlydlut/C2F-DFT/blob/main/Figs/fine.png#pic_center">

Requirements

Training and Evaluation

Training and testing instructions and visualization results for Image Deraining, Image Deblurring, and Real Image Denoising are provied in the links below.

<table> <tr> <th align="center">Task</th> <th align="center">Training</th> <th align="center">Testing</th> <th align="center">C2F-DFT's Visual Results</th> </tr> <tr> <td align="center">Image Deraining</td> <td align="center"><a href="Deraining/README.md#training">Link</a></td> <td align="center"><a href="Deraining/README.md#testing">Link</a></td> <td align="center"><a href="https://drive.google.com/drive/folders/1v4aAFDAojHtedtRmPcqVKJcAixW5dZ8m">Download</a></td> </tr> <tr> <td align="center">Image Deblurring</td> <td align="center"><a href="Deblurring/README.md#training">Link</a></td> <td align="center"><a href="Deblurring/README.md#testing">Link</a></td> <td align="center"><a href="https://drive.google.com/drive/folders/1qYVPblP0kCyfIoxDQ2NBsdbv_MoZ24S4">Download</a></td> </tr> <tr> <td align="center">Real Image Denoising</td> <td align="center"><a href="Denoising/README.md#training">Link</a></td> <td align="center"><a href="Denoising/README.md#testing">Link</a></td> <td align="center"><a href="https://drive.google.com/drive/folders/1hgSYcwSLktFh42LA9bDXTLUuNzThdJVA">Download</a></td> </tr> </table>

Experimental Results

<details> <summary><strong>Image Deraining</strong> (click to expand) </summary> <p align="center"><img src = "https://github.com/wlydlut/C2F-DFT/blob/main/Figs/tab1.png#pic_center"></p> <p align="center"><img src = "https://github.com/wlydlut/C2F-DFT/blob/main/Figs/fig3.png#pic_center" width="1000"></p> </details> <details> <summary><strong>Image Deblurring</strong> (click to expand) </summary> <p align="center"><img src = "https://github.com/wlydlut/C2F-DFT/blob/main/Figs/tab2.png#pic_center" width="500"></p> <p align="center"><img src = "https://github.com/wlydlut/C2F-DFT/blob/main/Figs/fig4.png#pic_center" width="1000"></p> </details> <details> <summary><strong>Real Image Denoising</strong> (click to expand) </summary> <p align="center"><img src = "https://github.com/wlydlut/C2F-DFT/blob/main/Figs/tab3.png#pic_center" width="500"></p> <p align="center"><img src = "https://github.com/wlydlut/C2F-DFT/blob/main/Figs/fig5.png#pic_center" width="1000"></p> </details>

Citation

@article{WANG2024104118,
title = {Coarse-to-fine mechanisms mitigate diffusion limitations on image restoration},
journal = {Computer Vision and Image Understanding},
volume = {248},
pages = {104118},
year = {2024},
issn = {1077-3142},
}

Acknowledgments

This code is based on the BasicSR toolbox and Restormer, WeatherDiffusion.