Matlab implementation for Deep Correlations for Texture Synthesis.
This package includes our implementation for Deep Correlations for Texture Synthesis. We did our best to make it a stande-alone package, meaning that we included ALL of the dependencies into this repository. The code was written by Omry Sendik.
Change the filename of the desired input texture in line 20 of Synth.m to reflect your desired input. For example:
origSrcImg = imread(strrep('.\Data\Texture13.png','\',filesep));
Note that the choice of Hyper-Parameters as described in the paper can be tweaked through GetSynthParams.m
Now, patiently wait for the result to converge. If things are working properly, you should get a plot similar to this:
./Data/Output/
Download our dataset from this repository too [data]
If you use this code for your research, please cite our paper:
@article{sendik2017deep,
title={Deep correlations for texture synthesis},
author={Sendik, Omry and Cohen-Or, Daniel},
journal={ACM Transactions on Graphics (TOG)},
volume={36},
number={5},
pages={161},
year={2017},
publisher={ACM}
}
Code relies heavily on:
MatConvNet
L-BFGS-B