Growing Neural Gas

Example video

This video is the compressed output of the GNG trained with the apple sample image in 2231 iterations with the default traits and at 100 fps.

Custom options

... or use the default options.

Set max iterations maxIterations, line thickness lineThick or select if the output should be a video or image exportMP4 on line 43 of main.cpp. If video, select the fps.

Default:

int maxIterations = 1000
int lineThick = 3
bool exportMP4 = true
int fps = 50

Set custom traits for the GNG on line 28 of main.cpp.

Default trait:

// Move winners by fractions of ..
float eb = 0.05;
float en = 0.03;

int maxAge = 10;
int lambda = 10; // Add point each .. iterations

// Decrease local error in each iteration by factors..
float alpha = 0.01;
float beta = 0.02;

Run

Without Docker

Requirement: OpenCV with ffmpeg

mkdir build
cd build
cmake ..
make

mkdir output
./gng ../sample-imgs/apple.jpg ../output

With Docker

Make sure output and sample-imgs are set as volumes to be able to watch the output and test new images.

docker build --tag alpine-gng .
mkdir output
docker run -itd --name gng -v $PWD/output:/gng/output -v $PWD/sample-imgs:/gng/sample-imgs alpine-gng
docker exec -it gng build/gng sample-imgs/apple.jpg output