https://github.com/nachifur/RDDM
CVPR 2024: Residual Denoising Diffusion Models
https://github.com/nachifur/RDDM
Last synced: 7 months ago
JSON representation
CVPR 2024: Residual Denoising Diffusion Models
- Host: GitHub
- URL: https://github.com/nachifur/RDDM
- Owner: nachifur
- Created: 2023-08-25T14:21:44.000Z (about 2 years ago)
- Default Branch: main
- Last Pushed: 2024-10-16T03:34:41.000Z (about 1 year ago)
- Last Synced: 2024-10-17T17:47:10.251Z (about 1 year ago)
- Language: Python
- Homepage: https://arxiv.org/abs/2308.13712
- Size: 9.05 MB
- Stars: 337
- Watchers: 2
- Forks: 32
- Open Issues: 2
-
Metadata Files:
- Readme: README.md
Awesome Lists containing this project
- awesome-diffusion-categorized - [Code
README
# Residual Denoising Diffusion Models
[paper](https://openaccess.thecvf.com/content/CVPR2024/html/Liu_Residual_Denoising_Diffusion_Models_CVPR_2024_paper.html)|[arxiv](https://arxiv.org/abs/2308.13712)|[youtube](https://www.youtube.com/watch?v=E-ObZs32fEU)|[blog](https://twitter.com/nachifur/status/1762730191707881537)|[中文论文(ao9l)](https://rec.ustc.edu.cn/share/60cb4770-1b6a-11ef-8e9e-332aeb6c199a)|[中文视频](https://cmdr.com.cn/lectureHall/lectureRoomDetail?liveUid=58e63bb51116d7c01f37dfee1354b043)|[中文博客](https://www.zhihu.com/question/645935461/answer/3410873004)
This repository is the official implementation of Residual Denoising Diffusion Models.
## Requirements
To install requirements: ([If an error occurs, you may need to install the packages one by one](https://github.com/nachifur/RDDM/issues/41#issuecomment-2477808693).)
```
conda env create -f install.yaml
```
## Dataset
[Raindrop](https://github.com/rui1996/DeRaindrop) ([test-a for test](https://github.com/rui1996/DeRaindrop))
[GoPro](https://github.com/swz30/MPRNet/blob/main/Deblurring/Datasets/README.md)
[ISTD](https://github.com/DeepInsight-PCALab/ST-CGAN)
SID-RGB: [kexu](https://kkbless.github.io/) or [download](https://drive.google.com/drive/folders/1-psXDjeW4FiRdLjc9idABsxGPo1Kn1jR)
[LOL](https://daooshee.github.io/BMVC2018website/)
[CelebA](https://github.com/nachifur/RDDM/issues/8#issuecomment-1978889073)
## Training
To train RDDM, run this command:
```train
cd experiments/xxxx
python train.py
```
or
```train
accelerate launch train.py
```
## Evaluation
To evaluate image generation, run:
```eval
cd eval/image_generation_eval/
python fid_and_inception_score.py path_of_gen_img
```
For image restoration, MATLAB evaluation codes in `./eval`.
## Pre-trained Models
[The pre-trained models (two unets, deresidual+denoising)](https://rec.ustc.edu.cn/share/3d8d9200-4e7e-11ef-b0ee-250e7e41f368) for [partially path-independent generation process](https://github.com/nachifur/RDDM/tree/main/experiments/0_Partially_path-independent_generation).
## Results
See Table 3 in main paper.
**For image restoration:**
[Raindrop](https://rec.ustc.edu.cn/share/c20ea640-4e7e-11ef-b29e-b1b12149494a)
[GoPro](https://rec.ustc.edu.cn/share/f9deffc0-4e7e-11ef-b4dd-b51790f24839)
[ISTD](https://rec.ustc.edu.cn/share/da867b10-4e7e-11ef-b21d-b3131e611f52)
[LOL](https://rec.ustc.edu.cn/share/e9c00ab0-4e7e-11ef-89a0-292c4c37c153)
[SID-RGB](https://rec.ustc.edu.cn/share/b213c330-4e7e-11ef-9b3e-957f50ca7d9b)
**For image generation (on the CelebA dataset):**
We can convert a pre-trained DDIM to RDDM by coefficient transformation (see [1_Image_Generation_convert_pretrained_DDIM_to_RDDM](https://github.com/nachifur/RDDM/tree/main/experiments/1_Image_Generation_convert_pretrained_DDIM_to_RDDM)).
## Citation
If you find our work useful in your research, please consider citing:
```
@InProceedings{Liu_2024_CVPR,
author = {Liu, Jiawei and Wang, Qiang and Fan, Huijie and Wang, Yinong and Tang, Yandong and Qu, Liangqiong},
title = {Residual Denoising Diffusion Models},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2024},
pages = {2773-2783}
}
```
## Contact
Please contact Jiawei Liu (liujiawei18@mails.ucas.ac.cn) or Liangqiong Qu (https://liangqiong.github.io/) if there is any question.