StarIR: Convolutional Image Restoration with Spatial-Frequency Fusion

:star: If StarIR is helpful for you, please help star this repo. Thanks! :hugs:

:new: News

• 2026.04.09: :fire: This repo is released

⚙️ Setup

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

:wrench: Training and Evaluation

Training and Testing instructions for Deraining, Dehazing, Desnowing, Motion deblurring, Defocus deblurring, Low-light enhancement, Low-light deblurring, and all-in-one tasks are provided in their respective directories. Here is a summary table containing hyperlinkes for easy navigation.

Task	Instructions	StarIR's Visual Results	Pre-trained Models	Datasets
Deraining	Link	Google Drive, 百度网盘	Google Drive, 百度网盘	Google Drive, 百度网盘
Dehazing	Link	Google Drive, 百度网盘	Google Drive, 百度网盘	Google Drive, 百度网盘
Desnowing	Link	Google Drive, 百度网盘	Google Drive, 百度网盘	Google Drive, 百度网盘
Motion deblurring	Link	Google Drive, 百度网盘	Google Drive, 百度网盘	Google Drive, 百度网盘
Defocus deblurring	Link	Google Drive, 百度网盘	Google Drive, 百度网盘	Google Drive, 百度网盘
Low-light enhancement	Link	Google Drive, 百度网盘	Google Drive, 百度网盘	Google Drive, 百度网盘
Low-light deblurring	Link	Google Drive, 百度网盘	Google Drive, 百度网盘	Google Drive, 百度网盘
All-in-one	Link	Google Drive, 百度网盘	Google Drive, 百度网盘	All_in_One, 百度网盘
UHD	[Link]	[Google Drive], 百度网盘	Google Drive, 百度网盘
CDD	[Link]	[Google Drive], 百度网盘	Google Drive, 百度网盘

💫 Demo

To test the pre-trained StarIR models of Single_Image_Defocus_Deblurring, Dehazing, Deraining, Desnowing, Low_Light_Deblurring, Low_Light_Enhancement, and Motion_Deblurring on your own images, you can download the models, place them in the /pretrained_models/ of the corresponding tasks (e.g., ./Motion_Deblurring/pretrained_models/StarIR-GOPRO.pth), then run

python demo.py --task Task_Name --input_dir path_to_images --result_dir save_images_here

Example usage to perform Defocus Deblurring on a directory of images:

python demo.py --task Single_Image_Defocus_Deblurring --input_dir './demo/degraded/' --result_dir './demo/restored/'

Example usage to perform Defocus Deblurring on an image directly:

python demo.py --task Single_Image_Defocus_Deblurring --input_dir './demo/degraded/portrait.jpg' --result_dir './demo/restored/'

For all-in-one demo, please refer to All_in_One

:notebook: Citation

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

@article{cui2026starir,
  title={StarIR: Convolutional Image Restoration With Spatial-Frequency Fusion},
  author={Cui, Yuning and Zamir, Syed Waqas and Yang, Ming-Hsuan and Knoll, Alois and Khan, Fahad Shahbaz and Khan, Salman},
  journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
  year={2026}
}

:envelope: Contact

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