PDE-Net

This is an implementation of Deep Posterior Distribution-based Embedding for Hyperspectral Image Super-resolution.

[arXiv], [IEEE]

Requirement

python 3.7, Pytorch 1.7.0, and cuda 11.0

Quick Start

Dataset

You can refer to the following links to download the datasets, CAVE, and Harvard. And run the matlab programs in the folder 'datasets' to get the pre-processed training and testing data.

Training

You can train directly by using the file 'train.sh':

bash train.sh

Or you can execute the following commands respectively:

python train.py --cuda --gpu "0" --dataset "CAVE" --upscale_factor 4 --model_name "template" --nEpochs 50

python train.py --cuda --gpu "0" --dataset "CAVE" --upscale_factor 4 --model_name "pde-net" --nEpochs 100 --resume checkpoints/CAVE_x4/template_4_epoch_50.pth

Testing

python test.py --cuda --gpu "0" --dataset "CAVE" --model_name "pde-net" --upscale_factor 4 --checkpoint checkpoints/CAVE_x4/pde-net_4_epoch_100.pth

Citation

Please consider cite our work if you find it helpful.

@article{hou22deep,
	title={Deep Posterior Distribution-based Embedding for Hyperspectral Image Super-resolution},
	author={Hou, Jinhui and Zhu, Zhiyu and Hou, Junhui and Zeng, Huanqiang and Wu, Jinjian and Zhou, Jiantao},
	journal={IEEE Transactions on Image Processing},
	volume={31},
	number={},
	pages={5720-5732},
	year={2022},
	doi={10.1109/TIP.2022.3201478}
}