Fig1. The Block Diagram of Video Snapshot Compressive Imaging
Video Snapshot Compressive Imaging (Video SCI) mainly focus on capturing highspeed scenario using a low speed camera and capturing high dimensional data in a single shot in order to minimize network transmission bandwidth, lowering time and hardware cost. These goals can be achieved by capturing 3D video data using a 2D detector and then storing and transmitting the video data into a 2D data matrix.
CACTI (Coded Aperture Compressive Temporal Imaging) is an classical Video SCI system. This library is implemented based on CACTI reconstruction algorithm via PyTorch. This library basically include all mainstream Video SCI algorithm.
CACTI Library was born in SCI Lab, Westlake University. We sincerely hope our library users can make their own contributions in various Video SCI algorithms, and by implementing and referring this library, one can broaden the library users and share deep understanding about Video SCI.
Fig2. Reconstructed Gray Value Data via Different Algorithms
Fig3. Reconstructed Colored Data via Different Algorithms
Fig3. Reconstruction Quality Comparison between Different Algorithms on Testing Dataset
All model parameters can be download at Dropbox and BaiduNetdisk
Please see the Installation Manual for CACTI Installation
- CACTI Code Library Documentation
- Model Training Dataset
- Newly Added Self-Defined Model
- Statistics of Model Params and FLOPs
- Images to Video and Images to GIF Transfer
- pytorch to onnx and onnx to tensorrt Tansform
Many appreciate to all the contributors who share their work for this library. Video SCI library always open and seeking for more Video SCI algorithms.
@article{2021Snapshot,
title={Snapshot Compressive Imaging: Principle, Implementation, Theory, Algorithms and Applications},
author={ Yuan, X. and Brady, D. J. and Katsaggelos, A. K. },
journal={IEEE Signal Processing Magazine},
year={2021},
}