azouaoui-cv/vae_synthetic_hsi

Sources for the method proposed in "Synthetic Hyperspectral Images with Controllable Spectral Variability and Ground Truth"

Python sources for the method proposed in

Synthetic Hyperspectral Images with Controllable Spectral Variability and Ground Truth

The code uses the Pytorch and Pytorch Lightning frameworks. The requirements needed to run the code is in the file requirements.txt. The method requires the hyperspectral images to be in Matlab mat files having the following named variables:

Variable	Content
Y	Array having dimensions B x P containing the spectra
GT	Array having dimensions R x B containing the reference endmembers
cols	The number of columns in the hyperspectral image (HSI)
rows	The number of rows in the HSI

Here, R is the number of endmembers, B the number of bands, and P the number of pixels. Edit the beta_vae.yaml file under the configs directory to change parameters for the beta_vae. Run the file train_vae.py to train the model. Edit and run the file run.py to generate synthetic HSIs. If you use the sources provided, make sure you cite the paper.