Multi-Modal Spectral Image Super-Resolution

Fayez Lahoud*, Ruofan Zhou*, Sabine Süsstrunk
ECCV Workshop 2018 Winner of [PIRM2018 Hyperspectral reconstruction challenge].

Citation

@inproceedings{lahoud2018multi,
  title={Multi-modal Spectral Image Super-Resolution},
  author={Lahoud, Fayez and Zhou, Ruofan and S{\"u}sstrunk, Sabine},
  booktitle={European Conference on Computer Vision},
  pages={35--50},
  year={2018},
  organization={Springer}
}

Code

Dependencies

• Pytorch 0.4.0
• cuDNN

Our code is tested under Ubuntu 14.04 environment with Titan X GPUs.

Inference for Track1

1. goto folder: data/track1/
2. run download_testing_data.sh
3. run generate_testing_h5.m
4. goto folder: code/track1/
5. run: python validate.py
6. run: python npz2mat.py
7. run mat2fla.m

the reconstruction is in code/track1/validation/*.{hdr,fla}

Inference for Track2

1. goto folder: data/track2/
2. run download_testing_data.sh
3. run generate_testing_h5.m
4. goto folder: code/track2/
5. run: python validate.py
6. run: python npz2mat.py
7. run mat2fla.m

the reconstruction is in code/track2/validation/*.{hdr,fla}

Training for Stage-I (Track1):

1. goto folder: data/track1/
2. run download_training_data.sh
3. run generate_training_h5.m
4. goto folder: code/track1/
5. run: cp -r ../../data/track1/hd5 ./data
6. run: python main.py

Training for Stage-II (Track2):

1. train Stage-I
2. goto folder: data/track2/
3. run download_training_data.sh
4. run generate_stage_one_h5.m
5. run: python generate_stage_one_results.py
6. run: python npz2mat.py
7. run mat2flat.m
8. run generate_training_h5.m
9. goto folder: code/track2/
10. run: cp -r ../../data/track2/hd5 ./data
11. run: python main.py

Authors

• Fayez Lahoud [Home]
• Ruofan Zhou
• Sabine Süsstrunk

License

• For academic and non-commercial use only.