Code for the paper: "Hyperspectral Image Super-resolution via Deep Spatio-spectral Attention Convolutional Neural Networks", TNNLS, 2021.
Project page: https://liangjiandeng.github.io/Projects_Res/HSRnet_2021tnnls.html
The proposed network is trained on Python 3.7.4 with Tensorflow 1.14.0 + cuda10.0 + cudnn 7.6.5
HSRnet.py: contains the training and testing code. Since the traning and validation data are too big, they are not provided in the folder. But the method of data simulation is described in the paper. Moreover, the models pre-trained with CAVE data or Harvard data are available (see below), you can select which model directory you want to use.
models(cave): The model trained with CAVE data;
models(harvard): The model trained with Harvard data;
test_cave_demo.mat : One testing image from CAVE dataset (download link: https://www.dropbox.com/s/u1cr6lye6xqppv0/test_cave_demo.mat?dl=0)
test_harvard_demo.mat : One testing image from Harvard dataset (download link: https://www.dropbox.com/s/5a3s7cijqfkwx13/test_harvard_demo.mat?dl=0)
If you find this code helpful, please kindly cite:
BibTeX: @article{Hu2021Hyperspectral, title={Hyperspectral Image Super-resolution via Deep Spatio-spectral Attention Convolutional Neural Networks}, author={ Hu, Jin-Fan and Huang, Ting-Zhu and Deng, Liang-Jian and Jiang, Tai-Xiang and Vivone, Gemine and Chanussot, Jocelyn}, journal={IEEE Transactions on Neural Networks and Learning Systems}, year={2021}, publisher={IEEE} }