This is a repo for our work: "Improving Spectral Snapshot Reconstruction with Spatial-Spectral Rectification".

Our work has been accepted by CVPR, codes and results are coming soon (July or August).

The codes and pre-trained weights have been released. More details and instructions will be continuously updated.

Download pre-trained weights from Github or Baidu disk then put them into the corresponding folder "SSR/Checkpoint_pretrain/" folder as the following form:

|--Checkpoint_pretrain
    |--model_L.pkl
    |--model_M.pkl
    |--model_S.pkl

For testing pre-trained models, run the following command

python Test_pretrain.py

Then run 'cal_psnr_ssim.m' in Matlab to get the performance metrics.

The simulated and real results of SSR are available here.

Python>=3.6
scipy
numpy
einops

Download the cave dataset of MST series from Baidu diskcode:fo0q or here, put the dataset into the corresponding folder "SSR/CAVE_1024_28/" as the following form:

|--CAVE_1024_28
    |--scene1.mat
    |--scene2.mat
    ：
    |--scene205.mat
    |--train_list.txt

Then run the following command

cd SSR
python Train.py

Download the test dataset from here, put the dataset into the corresponding folder "SSR/Test_data/" as the following form:

|--Test_data
    |--scene01.mat
    |--scene02.mat
    ：
    |--scene10.mat
    |--test_list.txt

Then run the following command

cd SSR
python Test.py

Finally, run 'cal_psnr_ssim.m' in Matlab to get the performance metrics.

If this repo helps you, please consider citing our work:

@InProceedings{SSR,
    author    = {Zhang, Jiancheng and Zeng, Haijin and Chen, Yongyong and Yu, Dengxiu and Zhao, Yin-Ping},
    title     = {Improving Spectral Snapshot Reconstruction with Spectral-Spatial Rectification},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2024},
    pages     = {25817-25826}
}