rh12503 / triangula

Generate high-quality triangulated and polygonal art from images.
MIT License
3.85k stars 123 forks source link
art evolutionary-algorithms evolutionary-art generative-art genetic-algorithm go golang gui polygons triangles triangula

An iterative algorithm to generate high quality triangulated and polygonal art from images.

Test status Go Reference Go Report Card License: MIT Tweet

Triangula uses a modified genetic algorithm to triangulate or polygonate images. It works best with images smaller than 3000px and with fewer than 3000 points, typically producing an optimal result within a couple of minutes. For a full explanation of the algorithm, see this page in the wiki.

You can try the algorithm out in your browser here, but the desktop app will typically be 20-50x faster.

Install

GUI

Install the GUI from the releases page. The GUI uses Wails for its frontend.

If the app isn't running on Linux, go to the Permissions tab in the executable's properties and tick Allow executing file as program.

CLI

Install the CLI by running:

go get -u github.com/RH12503/Triangula-CLI/triangula

Your PATH variable also needs to include your go/bin directory, which is ~/go/bin on macOS, $GOPATH/bin on Linux, and c:\Go\bin on Windows.

Then run it using the command:

triangula run -img <path to image> -out <path to output JSON>

and when you're happy with its fitness, render a SVG:

triangula render -in <path to outputted JSON> -img <path to image> -out <path to output SVG> 

For more detailed instructions, including rendering PNGs with effects see this page.

Options

For almost all cases, only changing the number of points and leaving all other options with their default values will generate an optimal result.

Name Flag Default Usage
Points --points, -p 300 The number of points to use in the triangulation
Mutations --mutations, --mut, -m 2 The number of mutations to make
Variation --variation, -v 0.3 The variation each mutation causes
Population --population, --pop, --size 400 The population size in the algorithm
Cutoff --cutoff, --cut 5 The cutoff value of the algorithm
Cache --cache, -c 22 The cache size as a power of 2
Block --block, -b 5 The size of the blocks used when rendering
Threads --threads, -t 0 The number of threads to use or 0 to use all cores
Repetitions --reps, -r 500 The number of generations before saving to the output file (CLI only)

Examples of output

Triangulated

Polygonal

Community Examples

Comparison to esimov/triangle

esimov/triangle seems to be a similar project to Triangula that is also written in Go. However, the two appear to generate very different styles. One big advantage of triangle is that it generates an image almost instantaneously, while Triangula needs to run many iterations.

esimov/triangle results were taken from their Github repo, and Triangula's results were generated over 1-2 minutes. esimov/triangle Triangula

Difference from fogleman/primitive and gheshu/image_decompiler

A lot of people have commented about Triangula's similarities to these other algorithms. While all these algorithms are iterative algorithms, the main difference is that in the other algorithms triangles can overlap while Triangula generates a triangulation.

API

Simple example:

import imageData "github.com/RH12503/Triangula/image"

func main() {
    // Open and decode a PNG/JPEG
    file, err := os.Open("image.png")

    if err != nil {
          log.Fatal(err)
    }

    image, _, err := image.Decode(file)

    file.Close()

    if err != nil {
          log.Fatal(err)
    }

    img := imageData.ToData(image)

    pointFactory := func() normgeom.NormPointGroup {
          return (generator.RandomGenerator{}).Generate(200) // 200 points
    }

    evaluatorFactory := func(n int) evaluator.Evaluator {
          // 22 for the cache size and 5 for the block size
          // use PolygonsImageFunctions for polygons 
          return evaluator.NewParallel(fitness.TrianglesImageFunctions(imgData, 5, n), 22)
    }

    var mutator mutation.Method

    // 1% mutation rate and 30% variation
    mutator = mutation.NewGaussianMethod(0.01, 0.3)

    // 400 population size and 5 cutoff
    algo := algorithm.NewModifiedGenetic(pointFactory, 400, 5, evaluatorFactory, mutator)

    // Run the algorithm
    for {
          algo.Step()
          fmt.Println(algo.Stats().BestFitness)
    }
}

Contribute

Any contributions are welcome. Currently help is needed with:

Thank you to these contributors for helping to improve Triangula: