Lambda-net

This repository contains the coder for the paper: λ-net: Reconstruct Hyperspectral Images from a Snapshot Measurement (International Conference on Computer Vision 2019) by Xin Miao, Xin Yuan, Yunchen Pu, Vassilis Athitsos.

Training and testing data.

The traing and testing data used in our paper is from the paper " Sparse Recovery of Hyperspectral Signal from Natural RGB Images". It can be downlaed from http://icvl.cs.bgu.ac.il/hyperspectral/. The training scene should be put in the file "training_data/",the testing scenes should be in the file "testing_data/". The data we used and the mask can be downloaded from https://drive.google.com/open?id=1lJB9Ekif4fXQSL9fGzrsxgp5YFMb7FEu.

Usage

0. Download the repository

Requirements are tensorflow, numpy, scipy.

1. Run Lambda-net

Train the model via

main.py

The results for testing scene will be saved in 'result/model file/'. If you want to visualize the results after training the model, just run

visualize_results.py

And the PSNR value for the training and testing data can be found in

result/model file/pnsr.txt