This repo contains my master's project on single image hyperspectral image super-resolution.
Model FPNSR
is an implementation of my IFTC 2019 paper.
Model SSPSR
is an implementation of the TCI 2020 paper.
Model IPNSR
is an enhanced version of the original IFTC 2019 model. It uses image-level pyramid instead of feature-level pyramid and achieves better results.
Chikusei dataset, sr factor x4:
Methods | MPSNR | MSSIM | CC | ERGAS | SAM | RMSE |
---|---|---|---|---|---|---|
FPNSR | 40.1784 | 0.9400 | 0.9549 | 5.1113 | 2.3348 | 0.0116 |
SSPSR | 40.3612 | 0.9413 | 0.9565 | 4.9894 | 2.3527 | 0.0114 |
IPNSR | 40.4876 | 0.9462 | 0.9577 | 4.8866 | 2.3368 | 0.0112 |
(SSPSR/FPNSR results are from the original paper.)
- Python 3.6
- PyTorch >1.1 (if you don't need tensorboard, >0.4 is fine)
- numpy
- cv2
- tqdm
- scipy
- pandas
- skimage
You also need MATLAB to generate your dataset. Please refer to ./data_preparation
.
Training:
python train.py --data_dir YOUR_DATASET_PATH --dataset Chikusei
--model IPNSR --sr_factor 4 --gpus 0, 1
Testing:
python test.py --gpus 0
You need to specify test_data_dir
, model_name
and save_model_title
in the global settings of test.py
.
If use use the above models, please cite:
@InProceedings{10.1007/978-981-15-3341-9_5,
author="Sun, He
and Zhong, Zhiwei
and Zhai, Deming
and Liu, Xianming
and Jiang, Junjun",
title="Hyperspectral Image Super-Resolution Using Multi-scale Feature Pyramid Network",
booktitle="Digital TV and Wireless Multimedia Communication",
year="2020",
publisher="Springer Singapore",
address="Singapore",
pages="49--61",
isbn="978-981-15-3341-9"
}
@article{jiang2020learning,
author={J. {Jiang} and H. {Sun} and X. {Liu} and J. {Ma}},
journal={IEEE Transactions on Computational Imaging},
title={Learning Spatial-Spectral Prior for Super-Resolution of Hyperspectral Imagery},
year={2020},
volume={6},
number={},
pages={1082-1096},}