Skip to content
Motion flow and corresponding blurry image synthesis. Training data generation for "From Motion Blur to Motion Flow: A Deep Learning Solution for Removing Heterogeneous Motion Blur (CVPR'17)".
Branch: master
Clone or download
Latest commit e43da13 May 18, 2018
Permalink
Type Name Latest commit message Commit time
Failed to load latest commit information.
conv_opt add mex files for MacOS Feb 8, 2018
dataset/image/train
imgs add overview img and update readme Feb 8, 2018
utils
README.md Updated readme May 18, 2018
gen_blurry_img.m import code and example script. Feb 8, 2018
gen_mf.m import code and example script. Feb 8, 2018
gen_mf_para.m import code and example script. Feb 8, 2018
script_data_gen.m fixed a bug on "q_type" setting. May 18, 2018

README.md

motion-flow-syn

Motion flow and corresponding blurry image synthesis. This package is used to generate training data for the paper:

From Motion Blur to Motion Flow: a Deep Learning Solution for Removing Heterogeneous Motion Blur
Dong Gong, Jie Yang, Lingqiao Liu, Yanning Zhang, Ian Reid, Chunhua Shen, Anton van den Hengel, Qinfeng Shi.
In IEEE Conference on Computer Vision and Pattern Recognition (CVPR), 2017.

[Paper][Project]

  • If you use this code for your research, please cite our paper:
@InProceedings{gong2017blur2mf,
author = {Gong, Dong and Yang, Jie and Liu, Lingqiao and Zhang, Yanning and Reid, Ian and Shen, Chunhua and van den Hengel, Anton and Shi, Qinfeng},
title = {From Motion Blur to Motion Flow: A Deep Learning Solution for Removing Heterogeneous Motion Blur},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {July},
year = {2017}
}

Motion blur and motion vector

Usage

  • Run makemexfiles in conv_opt/. The compiled .mex files are included, this step thus may be not necessary.
  • Try the example in script_data_gen.m. Two example for parameter setting are in the script.

Motion flow simulation

  • Please refer the paper for technique details.

Note

  • More related resources (e.g. paper and code) can be found on the project page.
  • Partial code in conv_opt/ is based on the implementation for paper "J. Sun, W. Cao, Z. Xu, and J. Ponce. Learning a convolutional neural network for non-uniform motion blur removal. In CVPR, 2015."
  • Partial visualization code is from here.

Update: 18 May 2018 Create: 8 February 2018

You can’t perform that action at this time.