Home

Awesome

[ECCV 2024] Hierarchical Separable Video Transformer for Snapshot Compressive Imaging

Ping Wang, Yulun Zhang, Lishun Wang, Xin Yuan

Video SCI Reconstruction Task and Its Degradation Analysis

pipeline

Video SCI Reconstruction Architecture

arch

HiSViT (building block)

hisvit

Result

sumary

Preparation

data

Please download testdata from https://drive.google.com/drive/u/0/folders/175mGay-6FS_B4dDNTUkayzMbikHfnsit and download traindata from DAVIS2017 dataset (480p) https://davischallenge.org/davis2017/code.html

checkpoint

Pretrained models are available at https://drive.google.com/drive/u/0/folders/175mGay-6FS_B4dDNTUkayzMbikHfnsit

mask

Masks are available at https://drive.google.com/drive/u/0/folders/175mGay-6FS_B4dDNTUkayzMbikHfnsit

Testing

HiSViT-9 Testing on gray dataset

python test.py --color_channels 1 --blocks 9 --size '[256,256]' --test_weight_path './checkpoint/hisvit9_gray.pth' --test_data_path './data/testdata/gray_256'

HiSViT-13 Testing on gray dataset

python test.py --color_channels 1 --blocks 13 --size '[256,256]' --test_weight_path './checkpoint/hisvit13_gray.pth' --test_data_path './data/testdata/gray_256'

HiSViT-9 Testing on color dataset

python test.py --color_channels 3 --blocks 9 --size '[512,512]' --test_weight_path './checkpoint/hisvit9_color.pth' --test_data_path './data/testdata/color_512'

Citation

If you use HiSViT, please consider citing:

@inproceedings{wang2025hierarchical,
  title={Hierarchical Separable Video Transformer for Snapshot Compressive Imaging},
  author={Wang, Ping and Zhang, Yulun and Wang, Lishun and Yuan, Xin},
  booktitle={European Conference on Computer Vision},
  pages={104--122},
  year={2025},
  organization={Springer}
}