[CVPR 2026] UniRain: Unified Image Deraining with RAG-based Dataset Distillation and Multi-objective Reweighted Optimization
May 25, 2026 Β· View on GitHub
Qianfeng Yang 1,2, Qiyuan Guan 1,2, Xiang Chen 2, Jiyu Jin* 1, Guiyue Jin 1, Jiangxin Dong 2
Dalian Polytechnic University1, Nanjing University of Science and Technology2
ποΈ Welcome to visit our website (δΈζ³¨εΊε±θ§θ§ι’εηδΏ‘ζ―ζε‘εΉ³ε°) for low-level vision:https://lowlevelcv.com/
π οΈ Overview
Overall framework of UniRain. (Left) The RAG-based dataset distillation pipeline retrieves real rainy references consistent with the query image via multi-level similarity search and employs vision language models to evaluate its quality, thereby distilling reliable samples from public datasets. (Right) The asymmetric MoE architecture consists of soft-MoE encoder and hard-MoE decoder, optimized via the multi-objective reweighted strategy to achieve balanced learning and robust performance across multiple rain degradation types.
π© New Features/Updates
- β May 24, 2026. Release model code.
- β March 19, 2026. Release the training and testing code.
- β March 05, 2026. Release UniRain Paper.
- β March 03, 2026. Release the dataset, visual results, and testing code.
- β February 21, 2026. π Our UniRain was accepted by CVPR 2026!
π₯ RainRAG Dataset Download
| Mixed training dataset | Classified training dataset | Test datasets |
|---|---|---|
| Baidu Netdisk | Baidu Netdisk | Baidu Netdisk |
π Setup
Environment:
conda env create --name UniRain -f environment.yml
conda activate UniRain
How to train:
python train.py
How to evaluation:
python test.py
π Performance Evaluation
π·οΈ Visual Results
π Citation
If this work is helpful for your research, please consider citing the following BibTeX entry.
@article{UniRain,
title={UniRain: Unified Image Deraining with RAG-based Dataset Distillation and Multi-objective Reweighted Optimization},
author={Yang, Qainfeng and Guan, Qiyuan and Chen, Xiang and Jin, Jiyu and Jin, Guiyue and Dong, Jiangxin},
journal={CVPR},
year={2026}
}
π Contact
If you have any questions, please feel free to reach us out at csqianfengyang@163.com.