c-yn / chair

[Knowledge-Based Systems] Exploring the Potential of Channel Interactions for Image Restoration

Exploring the potential of channel interactions for image restoration

Yuning Cui, Alois Knoll

(Training details can be found in the supplementary material, which can also be downloaded from this link.)

Image restoration aims to reconstruct a clear image from a degraded observation. Convolutional neural networks have achieved promising performance on this task. The usage of Transformer has recently made significant advancements in state-of-the-art performance by modeling long-range dependencies. However, these deep architectures primarily concentrate on enhancing representation learning for the spatial dimension, neglecting the significance of channel interactions. In this paper, we explore the potential of channel interactions for restoring images through our proposal of a dual-domain channel attention mechanism. To be specific, channel attention in the spatial domain allows each channel to amass valuable signals from adjacent channels under the guidance of learned dynamic weights. In order to effectively exploit the significant difference in infrequency between degraded and clean image pairs, we develop the implicit frequency domain channel attention to facilitate the integration of information from different frequencies. Extensive experiments demonstrate that the proposed network, dubbed ChaIR, achieves state-of-the-art performance on 13 benchmark datasets for five image restoration tasks, including image dehazing, image motion/defocus deblurring, image desnowing, and image deraining.

Installation

The project is built with PyTorch 3.8, PyTorch 1.8.1. CUDA 10.2, cuDNN 7.6.5 For installing, follow these instructions:

conda install pytorch=1.8.1 torchvision=0.9.1 -c pytorch
pip install tensorboard einops scikit-image pytorch_msssim opencv-python

Install warmup scheduler:

cd pytorch-gradual-warmup-lr/
python setup.py install
cd ..

Training and Evaluation

Please refer to respective directories.

Results (ChaIR)

Images here

Task	Dataset	PSNR	SSIM
Motion Deblurring	GoPro	33.28	0.963
	HIDE	30.97	0.941
	RSBlur	34.25	0.871
Image Dehazing	SOTS-Indoor	41.95	0.997
	SOTS-Outdoor	40.73	0.997
	Dense-Haze	17.50	0.62
	NHR	28.18	0.98
Image Desnowing	CSD	39.24	0.99
	SRRS	31.91	0.98
	Snow100K	33.79	0.95
Image Deraining	Rain100L	38.20	0.973
	Rain100H	31.74	0.906
Defocus Deblurring	DPDD	26.29	0.816

Citation

If you find this project useful for your research, please consider citing:

@article{cui2023exploring,
  title={Exploring the potential of channel interactions for image restoration},
  author={Cui, Yuning and Knoll, Alois},
  journal={Knowledge-Based Systems},
  volume={282},
  pages={111156},
  year={2023},
  publisher={Elsevier}
}

Contact

Should you have any question, please contact Yuning Cui.

chair's Issues

关于复现

老师您好，这里面的数据集地址该怎么填写呢，根据您提供的here里面的链接下载下来之后怎么进一步复现呢？

开源

请问什么时候开源代码和预训练模型呢

fantastic work !when the code will be released？

fantastic work !when the code will be released？

About the DenseHaze dataset.

When training with the DenseHaze dataset, what size patches do you crop the images into? 256×256 or 800×1200?