Exploring the Potential of Pooling Techniques for Universal Image Restoration [TIP-25]

News

20/05/2025 We release visual results and pre-trained models for dehazing, desnowing, lolblur, cdd, and all-in-one tasks.

You are welcome to apply our lightweight plugin module to other tasks.

Installation

See Install

Training and Evaluation

Please refer to the respective directories.

Visual Results: gdrive

Models: gdrive

Complexity

Base model: 66.38 GFLOPs, Parameters 6.92M

Small model: 37.3 GFLOPs, Parameters 3.82M

Citation

@article{poolnet,
  title={Exploring the Potential of Pooling Techniques for Universal Image Restoration},
  author={Cui, Yuning and Ren, Wenqi and Knoll, Alois},
  journal={IEEE Transactions on Image Processing},
  year={2025},
  publisher={IEEE}
}