HSGAN

Supplementary material for "HSGAN: Hyperspectral Reconstruction from RGB Images with Generative Adversarial Network", accepted by IEEE Transactions on Neural Networks and Learning Systems, 2023

1 Preparation

We test the code on CUDA 8.0 (higher version is also compatible). The basic requirements are as follows:

• pytorch==1.0.0
• torchvision==0.2.1
• h5py
• python-opencv
• spectral
• scikit-image

2 Use the code

• Train a model:

python train.py

The default method is trainer.Trainer for the first 3000 training epochs. Please then use trainer.Trainer_GAN method after training with 3000 epochs.

• Validate a trained model:

python validation.py

3 Division of training / validation sets

The following table concludes the division of training and validation sets:

4 Citation

If you think this work is helpful, please consider cite:

@article{zhao2023hsgan,
  title={HSGAN: Hyperspectral Reconstruction from RGB Images with Generative Adversarial Network},
  author={Yuzhi Zhao, Lai-Man Po, Tingyu Lin, Qiong Yan, Wei Liu, Pengfei Xian},
  journal={IEEE Transactions on Neural Networks and Learning Systems (TNNLS)},
  year={2023}
}