denghong-liu/EUNet

A Pytorch implementation of "An Efficient Unfolding Network with Disentangled Spatial-Spectral Representation for Hyperspectral Image Super-Resolution"

An Efficient Unfolding Network with Disentangled Spatial-Spectral Representation for Hyperspectral Image Super-Resolution

Pytorch implementation of EUNet [Paper]

Requirements

• Python 3.9
• Pytorch 1.12.1

Training

To train EUNet, run the following commands. You may need to change the dir_data, dataset_name, scale, n_colors, is_blur, learning_rate, etc. in the option.py file for different settings.

# Bicubic downsampling
python main.py --scale 2 --dir_data hdata/data/ --dataset_name Pavia  --n_colors 102

# Gaussian downsampling
python main.py --scale 2 --dir_data hdata/data/ --dataset_name Pavia  --n_colors 102 --is_blur True --learning_rate 1e-3

Testing

For your convience, we provide the testset of Pavia Centre in /hdata/data/ and the pretrained 2X model in /hsr/model/.

python main_test.py --scale 2 --dir_data hdata/data/ --dataset_name Pavia  --n_colors 102 --model_path hsr/model/G.pth

Citation

Please cite our work in your publications if it helps your research.

@article{liu2023efficient,
  title={An Efficient Unfolding Network with Disentangled Spatial-Spectral Representation for Hyperspectral Image Super-Resolution},
  author={Liu, Denghong and Li, Jie and Yuan, Qiangqiang and Zheng, Li and He, Jiang and Zhao, Shuheng and Xiao, Yi},
  journal={Information Fusion},
  year={2023},
  publisher={Elsevier},
  doi={10.1016/j.inffus.2023.01.018}
}

References

• KAIR