If you find this code is useful, please cite our paper:

% @article{liu2019convergence,
% title={On the convergence of learning-based iterative methods for nonconvex inverse problems},
% author={Liu, Risheng and Cheng, Shichao and He, Yi and Fan, Xin and Lin, Zhouchen and Luo, Zhongxuan},
% journal={IEEE transactions on pattern analysis and machine intelligence},
% year={2019},
% publisher={IEEE}
% }

We provide a compiled Matconvnet-1.0-beta24 in Windows10, CUDA9.0, GTX TITAN X, but in most cases, you need to recompile it in you own machine with vl_compilenn() function. Other version of Matconvnet may also work.

Make sure your Matconvnet is compiled and its reference path is set correctly.

Simplely run Blind_Deblur.m, Nonblind_Deblur.m or Derain.m in Matlab, you can see the results shown below. It's quite easy!

|| | |:----------:|:-------------:| | Input | Output |

|| | |:----------:|:-------------:| | GT Kernel: size = 75 | Our Estimated Kernel |

|| | |:----------:|:-------------:| | Input | Output |

|| | |:----------:|:-------------:| | Input | Output |

|| | |:----------:|:-------------:| | Input | Output |