Skip to content
C++ implementation of a networks converter from JPG images.
Branch: master
Clone or download
Fetching latest commit…
Cannot retrieve the latest commit at this time.
Type Name Latest commit message Commit time
Failed to load latest commit information.

Fast image to network generator

C++ implementation of a network converter from JPG images.

Other examples



The two required packages are boost::program_options and CImg.

cd build
cmake ../src

CImg can be tricky to import with cmake depending on your machine. The current CMakeLists.txt has been tested on MacOSX 10.13.1. However, on Ubuntu, you may need to install the following packages:

apt-get install cmake libx11-dev libjpeg-dev cimg-dev gnuplot

Creating networks

After compiling, the executable GRGImage should have been created in the build folder. The parameters are passed using boost::program_options so you can do

./GRGImage --help

to obtain a complete list of the parameters.

List of parameters

Flag Type Description
--imagePath (-i) String Path to the JPG image
--outputPath (-o) String Path to dump the output data (add prefix of data output)
--number_nodes (-N) Int Number of nodes in the resulting network
--degree (-d) Int Maximum degree of a node (usually 3 or 4)
--constrast Float Contrast parameter between -infinity to infinity. If positive, darker pixels will have a larger weight than lighter pixels. If you have a dark background, I suggest to use negative constrats.
--neigh (-n) Int maximum number of neighbor pixels that a node can connect to (for faster convergence)
--help Show the parameters list
--transform Required to transform the network
--displayGray Not implemented
--CSV Output the data in CSV format
--GNU Output the data in a Gnuplot compatible format


Examples have been made available in the examples folder.

For the glasses example, we run:

../build/GRGImage -i ./images/glasses.jpg -o ./data/glasses -N 1000 -d 3 --contrast 3.7 --transform --CSV --GNU

1000 nodes are enough to get a sharp image.


A Gnuplot script does the plotting for us. In the past, I used matplotlib but it was very slow (2-3 minutes for plotting). I also used a d3.js script to show the images in the browser that I could release if requested.

The Gnuplot script takes some parameters

Params Type Description
dat String Path to the dat file
o String Output path of the image (do not include the extension)
w Int Width of the original image (written on the dat file)
h Int Height of the original image (written on the dat file)
r Float The ratio between the output size and the original size
png Int If png=1, the image will be png
pdf Int If pdf=1, the image will be pdf
lw Float Linewidth of the edges

An example is

gnuplot -e "dat='./data/glasses_width1200_height748.dat'; o='./networks/glasses';w=1200;h=748;r=1;png=1;lw=1.8" ../src/GNU_plot.gpi
You can’t perform that action at this time.