canbaba0517 / stdan

Implementation of "Spatio-Temporal Deformable Attention Network for Video Deblurring". (Zhang et al., ECCV 2022)

Spatio-Temporal Deformable Attention Network for Video Deblurring (ECCV2022)

PyTorch codes for "Spatio-Temporal Deformable Attention Network for Video Deblurring (ECCV2022)"

Datasets

We use the GoPro, DVD and BSD datasets in our experiments, which are available below:

• GoPro
• DVD
• BSD

Pretrained Models

You could download the pretrained model from here and put the weights in weights folder.

Prerequisites

Clone the Code Repository

git clone https://github.com/huicongzhang/STDAN.git

Install Pytorch Denpendencies

conda create -n STDAN python=3.7 
conda activate STDAN
conda install pytorch==1.10.1 torchvision==0.11.2 torchaudio==0.10.1 cudatoolkit=10.2 -c pytorch

Install Python Denpendencies and Build PyTorch Extensions

cd STDAN
sh install.sh

Get Started

To train STDAN, you can simply use the following command:

python runner.py --data_path=/yourpath/DeepVideoDeblurring_Dataset/quantitative_datasets --data_name=DVD --phase=train

To test STDAN, you can simply use the following command:

python runner.py --data_path=/yourpath/DeepVideoDeblurring_Dataset/quantitative_datasets --data_name=DVD --phase=test --weights=./weights/DVD_release.pth 

In here, there are more settings of testing and training.

Some video results are shown in here

Cite this work

@inproceedings{zhang2022spatio,
    title={Spatio-Temporal Deformable Attention Network for Video Deblurring},
    author={Zhang, Huicong and Xie, Haozhe and Yao, Hongxun},
    booktitle={ECCV},
    year={2022}
}

License

This project is open sourced under MIT license.

Acknowledgement

This project is based on STFAN