This repo is the implementation of "Spatial-Temporal Transformer for Video Snapshot Compressive Imaging".
Video snapshot compressive imaging (SCI) captures multiple sequential video frames by a single measurement using the idea of computational imaging. The underlying principle is to modulate high-speed frames through different masks and these modulated frames are summed to a single measurement captured by a low-speed 2D sensor (dubbed optical encoder); following this, algorithms are employed to reconstruct the desired high-speed frames (dubbed software decoder) if needed. In this paper, we consider the reconstruction algorithm in video SCI, i.e., recovering a series of video frames from a compressed measurement. Specifically, we propose a Spatial-Temporal transFormer (STFormer) to exploit the correlation in both spatial and temporal domains. STFormer network is composed of a token generation block, a video reconstruction block, and these two blocks are connected by a series of STFormer blocks. Each STFormer block consists of a spatial self-attention branch, a temporal self-attention branch and the outputs of these two branches are integrated by a fusion network. Extensive results on both simulated and real data demonstrate the state-of-the-art performance of STFormer.
Fig1. Reconstructed Color Data via Different Algorithms
Please see the Installation Manual for STFormer Installation.
Support multi GPUs and single GPU training efficiently. First download DAVIS 2017 dataset from DAVIS website, then modify data_root value in configs/_base_/davis.py file, make sure data_root link to your training dataset path.
Launch multi GPU training by the statement below:
CUDA_VISIBLE_DEVICES=0,1,2,3 python -m torch.distributed.launch --nproc_per_node=4 --master_port=3278 tools/train.py configs/STFormer/stformer_base.py --distributed=True
Launch single GPU training by the statement below.
Default using GPU 0. One can also choosing GPUs by specify CUDA_VISIBLE_DEVICES
python tools/train.py configs/STFormer/stformer_base.py
Specify the path of weight parameters, then launch 6 benchmark test in grayscale simulation dataset by executing the statement below.
python tools/test.py configs/STFormer/stformer_base.py --weights=checkpoints/stformer_base.pth
First, download the model weight file (checkpoints/stformer/stformer_base_mid_color.pth) and test data (datasets/middle_scale) from Dropbox or BaiduNetdisk, and place them in the checkpoints folder and test_datasets folder respectively. Then, execute the statement below to launch STFormer in 6 middle color simulation dataset.
python tools/test.py configs/STFormer/stformer_base_mid_color.py --weights=checkpoints/stformer_base_mid_color.pth
Download model weight file (checkpoints/stformer/stformer_base_real_cr10.pth) from Dropbox or BaiduNetdisk. Launch STFormer on real dataset by executing the statement below.
python tools/test_real_data.py configs/STFormer/stformer_base_real_cr10.py --weights=checkpoints/stformer_base_real_cr10.pth
Notice:
Results only show real data when its compress ratio (cr) equals to 10, for other compress ratio, we only need to change the cr value in file in stformer_real_cr10.py and retrain the model.
@article{wang2023spatial,
author={Wang, Lishun and Cao, Miao and Zhong, Yong and Yuan, Xin},
journal={IEEE Transactions on Pattern Analysis and Machine Intelligence},
title={Spatial-Temporal Transformer for Video Snapshot Compressive Imaging},
year={2023},
volume={45},
number={7},
pages={9072-9089},
doi={10.1109/TPAMI.2022.3225382}}
The codes are based on CACTI, we also refer to codes in Swin Transformer, Video Swin Transformer, RevSCI and Two Stage. Thanks for their awesome works.