README.md

SUNet: Swin Transformer with UNet for Image Denoising (ISCAS 2022)

Chi-Mao Fan, Tsung-Jung Liu, Kuan-Hsien Liu

Paper:

Video Presentation:

Presentation Slides:

---

Abstract : Image restoration is a challenging ill-posed problem which also has been a long-standing issue. In the past few years ago, the convolution neural networks (CNN) almost dominated the compute vision and had achieved considerable success in different level of vision tasks including image restoration. However, the Swin Transformer-based model also shows impressive performance, even suppress the CNN-based methods to become the state-of-the-art on high-level vision tasks recently. In this paper, we proposed a restoration model called SUNet which uses the Swin Transformer layer as our basic blocks and apply to U-Net architecture for image denoising.

Network Architecture

Overall Framework of SUNet
Swin Transformer Layer	Dual up-sample

Quick Run

To test the pre-trained models of denoising on your own images, run

python demo.py --input_dir images_folder_path --result_dir save_images_here --weights path_to_models

Here is an example to perform Deraindrop:

python demo.py --input_dir './demo_samples/' --result_dir './demo_results' --weights './pretrained_model/denoising_model.pth'

Result

Visual Comparison

More Visual Comparison Figures (Click to expand)

Citation

If you use SUNet, please consider citing:

@inproceedings{,
    title={},
    author={Chi-Mao Fan, Tsung-Jung Liu, Kuan-Hsien Liu},
    booktitle={ISCAS},
    year={2022}
}