Bio-Inspired Image Restoration

⭐ If BioIR is helpful for you, please help star this repo. Thanks! 🤗

🆕 News

• 2025.12.23: 🔥 This repo is released 🎄 Merry Christmas! and Happy New Year!

⚙️ Setup

See INSTALL.md for the installation of dependencies required to run BioIR.

🔧 Training and Evaluation

Training and Testing instructions for single-degradation&composite degradation and all-in-one tasks are provided in Single_Composite and All_in_One directories, respectively.

Tasks	Instructions	Visual Results	Pre-trained Models	Datasets
Single and composite degradation	Link	百度网盘	Google Drive, 百度网盘	Desnowing (Google Drive, 百度网盘). Dehazing (Google Drive, 百度网盘). CDD(百度网盘). LOLBlur (百度网盘)
All-in-one	Link	Google Drive, 百度网盘	Google Drive, 百度网盘	百度网盘

💫 Demo

(For all-in-one demo, please refer to demo.py.) To test the pre-trained BioIR models on your own images, you can download the models, place them in pretrain_model.

Example: use the model pretrained on CDD on your own images:

python demo.py --input_dir './demo/degraded/' --result_dir './demo/restored/' --dataset CDD

python demo.py --input_dir './demo/degraded/1.png' --result_dir './demo/restored/' --dataset CDD

📓 Citation

Please cite us if our work is useful for your research.

@inproceedings{bioir,
title={Bio-Inspired Image Restoration},
author={Yuning Cui and Wenqi Ren and Alois Knoll},
booktitle={The Thirty-ninth Annual Conference on Neural Information Processing Systems},
year={2025}
}

✉️ Contact

Should you have any question, please contact yuning.cui@in.tum.de