Rethinking Nighttime Image Deraining via Learnable Color Space Transformation (NeurIPS 2025)

Qiyuan Guan* ¹, Xiang Chen* ², Guiyue Jin ¹, Jiyu Jin ¹, Shumin Fan ³, Tianyu Song ³, Jinshan Pan ²

Dalian Polytechnic University¹, Nanjing University of Science and Technology², Dalian Martime University³

[Paper]

👉️ Welcome to visit our website (专注底层视觉领域的信息服务平台) for low-level vision: https://lowlevelcv.com/

---

⛳️ To do

• ✅ Release the code
• ✅ Release the visual results
• ✅ Release the dataset

---

⬇️ HQ-NightRain Dataset Download

Download Link
Google Drive / Baidu Netdisk (asht)

---

🛠 Setup

• Type the command:

conda env create -f environment.yml
conda activate CSTNet

• If torch1.10 download fails, please run the following command：

conda install pytorch==1.10.0 torchvision==0.11.0 torchaudio==0.10.0 cudatoolkit=11.3 -c pytorch -c conda-forge

📷️ Visual Results

● Comparative method results

Datasets	Visual Results Download Link
HQ-NightRain-RS	Google Drive / Baidu Netdisk (adai)
HQ-NightRain-RD	Google Drive / Baidu Netdisk (m1kb)
HQ-NightRain-SD	Google Drive / Baidu Netdisk (k7aq)
GTAV-NightRain	Google Drive / Baidu Netdisk (b6sm)
RealRain1k-L	Google Drive / Baidu Netdisk (zi73)
RealRain1k-H	Google Drive / Baidu Netdisk (r4cr)
RainDS-real-RS	Google Drive / Baidu Netdisk (98bt)
RainDS-real-RD	Google Drive / Baidu Netdisk (1ed3)
RainDS-real-RDS	Google Drive / Baidu Netdisk (9mhh)

● CST-Net results

Visual Results Download Link
Google Drive / Baidu Netdisk (n3vg)

Due to storage limitations, please contact us to obtain the Google Drive link.

---

🧮 Evaluation

● Run the following code to obtain the output visual results

python test.py

And you can find the output visual results in the folder " results/test/ ".

● Install the environment

We use the code provided by IQA-PyTorch for evaluation. Thanks to Chaofeng Chen!

pip install pyiqa

● Run the following command to calculate the metrics

python cal_metrics.py --inp_imgs ./results --gt_imgs ./dataset/test/target --log path_save_log

💪 Training

Run the following code to start training.

python train.py

👍 Acknowledgement

Thanks for their awesome works (IQA-PyTorch and NeRD-Rain)

---

❣ Citation

If this work is helpful for your research, please consider citing the following BibTeX entry.

@article{guan2025cstnet,
  title={Rethinking Nighttime Image Deraining via Learnable Color Space Transformation},
  author={Guan, Qiyuan and Chen, Xiang and Jin, Guiyue and Jin, Jiyu and Fan, Shumin and Song, Tianyu and Pan, Jinshan},
  journal={NeurIPS},
  year={2025}
}

📧 Contact

If you have any questions, please feel free to contact [email protected].

---