This repository contains the source code for the paper: SUNet: Symmetric Undistortion Network for Rolling Shutter Correction (ICCV2021). We can remove the jelly effect from the rolling shutter image on the left to produce the corresponding global shutter image on the right.

Install the dependent packages:

pip install -r requirements.txt

The code is tested with PyTorch 1.6.0 with CUDA 10.2.89.

Note that in our implementation, we borrowed some modules from DeepUnrollNet.

cd ./package_correlation
python setup.py install

cd ./package_forward_warp
python setup.py install

cd ./package_core
python setup.py install

You can now test our model with the provided images in the demo folder. To do this, simply run

sh demo.sh

The visualization results will be stored in the experiments folder. Other examples in the dataset can be tested similarly.

• Carla-RS and Fastec-RS: Download them to your local computer from here.

You can run following commands to re-train the network.

# !! Please update the corresponding paths in 'train.sh' with  #
# !! your own local paths, before run following command!!      #
sh train.sh

You can run following commands to obtain the quantitative evaluations.

# !! Please update the path to test data in 'inference.sh'
# !! with your own local path, before run following command!!
sh inference.sh

Please cite our paper if necessary:

@inproceedings{fan_SUNet_ICCV21,
  title={SUNet: Symmetric Undistortion Network for Rolling Shutter Correction},
  author={Fan, Bin and Dai, Yuchao and He, Mingyi},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={4541--4550},
  year={2021}
}

Please drop me an email for further problems or discussion: binfan@mail.nwpu.edu.cn