/HSCNN-Plus

HSCNN+: Advanced CNN-Based Hyperspectral Recovery from RGB Images in CVPRW 2018 (Winner of NTIRE Challenge)

Primary LanguagePython

HSCNN+: Advanced CNN-Based Hyperspectral Recovery from RGB Images

Zhan Shi, Chang Chen, Zhiwei Xiong, Dong Liu, Feng Wu. HSCNN+: Advanced CNN-Based Hyperspectral Recovery from RGB Images. In CVPR Workshop 2018. (Winner of NTIRE Challenge on Spectral Reconstruction from RGB Images)

Test the pre-trained models

Usage example to test HSCNN-D model for clean RGB images

cd hscnn-d_clean
cd inference/models && unzip *.zip && cd ../
/bin/bash demo.sh

Usage example to test HSCNN-D model for real-world RGB images

cd hscnn-d_real
cd inference/models && unzip *.zip
cat *.tar.gz.* | tar -xzv && cd ../
/bin/bash demo.sh

To access the validation dataset and the reconstructed results of HSCNN-D
Download the NTIRE2018_Validate folder from
http://pan.bitahub.com/index.php?mod=shares&
sid=eTJ2bFFQR3BzTm5FTGxjcC1WUWk3TXRsbGo3YTBjYi05SWVvSlE

Update 2019/06/05: Model-ensembled results of HSCNN-D are now replaced by the single-model ones (Table 3).

Train the model

Usage example to train a new model for clean RGB images

cd hscnn-d_clean/train && python train.py

Usage example to train a new model for real-world RGB images

cd hscnn-d_real/train && python train.py

Datasets

ICVL http://icvl.cs.bgu.ac.il/hyperspectral
NTIRE2018 http://icvl.cs.bgu.ac.il/ntire-2018