Bitmap color clustering by ML.NET

ClusterColors.cs

using Microsoft.ML;
using Microsoft.ML.Data;
using Microsoft.ML.Trainers;

using SkiaSharp;

public record ColorCount(
    byte R,
    byte G,
    byte B,
    int Count);

public sealed class RgbData
{
    public float R;
    public float G;
    public float B;
}

public sealed class ClusterPrediction
{
    [ColumnName("PredictedLabel")]
    public uint ClusterId;

    [ColumnName("Score")]
    public float[] Distances = default!;
}

public List<ColorCount> ClusterColors(
    SKBitmap bitmap,
    int maxClusters,
    int maxIterations,
    float tolerance)
{
    var width = bitmap.Width;
    var height = bitmap.Height;

    var colors = new HashSet<SKColor>();
    var pixels = new RgbData[width * height];
    var index = 0;
    for (var y = 0; y < height; y++)
    {
        for (var x = 0; x < width; x++)
        {
            var color = bitmap.GetPixel(x, y);
            pixels[index++] = new RgbData { R = color.Red, G = color.Green, B = color.Blue };

            if (colors.Count < maxClusters)
            {
                colors.Add(color);
            }
        }
    }

    var actualClusters = Math.Min(maxClusters, colors.Count);

    // KMeans
    var mlContext = new MLContext();
    var dataView = mlContext.Data.LoadFromEnumerable(pixels);

    var options = new KMeansTrainer.Options
    {
        FeatureColumnName = "Features",
        NumberOfClusters = actualClusters,
        MaximumNumberOfIterations = maxIterations,
        OptimizationTolerance = tolerance,
        //InitializationAlgorithm = KMeansTrainer.InitializationAlgorithm.KMeansPlusPlus
        InitializationAlgorithm = KMeansTrainer.InitializationAlgorithm.Random
    };
    var pipeline = mlContext.Transforms
        .Concatenate("Features", nameof(RgbData.R), nameof(RgbData.G), nameof(RgbData.B))
        .Append(mlContext.Clustering.Trainers.KMeans(options));

    var model = pipeline.Fit(dataView);
    var transformed = model.Transform(dataView);

    // Get center
    var centroids = default(VBuffer<float>[]);
    model.LastTransformer.Model.GetClusterCentroids(ref centroids, out _);

    // Count
    var counts = new int[actualClusters];
    foreach (var prediction in mlContext.Data.CreateEnumerable<ClusterPrediction>(transformed, reuseRowObject: false))
    {
        counts[prediction.ClusterId - 1]++;
    }

    var list = new List<ColorCount>(actualClusters);
    for (var i = 0; i < counts.Length; i++)
    {
        var centroid = centroids[i].DenseValues().ToArray();
        var r = (byte)Math.Round(centroid[0]);
        var g = (byte)Math.Round(centroid[1]);
        var b = (byte)Math.Round(centroid[2]);
        list.Add(new ColorCount(r, g, b, counts[i]));
    }

    list.Sort(static (x, y) => y.Count - x.Count);
    return list;
}