Hyperspectral unmixing task aims to address the complex spectral mixtures in hyperspectral data by decomposing each pixel’s spectral signature into pure spectral signatures ( endmembers ) and their corresponding proportions ( abundances ).
This repo contains the Urban dataset. You can download it here.
Predicting the abundance maps and the endmemebers by reconstructing the hyperpsectral patches with the tailored auto-encoder based model.
trainval.py
Note: 1) please download the pretrained checkpoint pth :
Spatial_MAE ViT-B;
Spectral_MAE ViT-B;
2) please download the hyperspectral unmixing dataset;
3) please put the pretrained model file and dataset in the file './data/';
Please see func.get_args for more details .
SpatSIGMA_Unmix
HyperSIGMA_Unmix
Figure. Framework of HyperSIGMA_Unmix.
This project is partly based on CNNAEU and DeepTrans.
Thanks for their wonderful work!