GetPixelsFast too slow for our use-case

[derwaldgeist]

Hi, we have an AR app that downloads a couple of textures for objects around the user. For these textures, we have to know the predominant color. The texture images (PNGs) are approx. 1.5mb.

We noticed significant frame drops, and when we analysed the root cause, we learned that ColorThief's "GetPixelFast" method is the actual culprit. More specifically, the following line:

var pixels = imageData.ToArray();

This array conversion takes around 1.5s. And it doesn't matter which quality setting we choose, as this array conversion is done before quality is taken into account.

Is there a way to speed this up, i.e. by copying only the required pixels into an array?

[derwaldgeist]

I was able to speed things up drastically:

private int[][] GetPixelsFast(Texture2D sourceImage, int quality, bool ignoreWhite, int mipLevel) {
		var mipData = sourceImage.GetPixels32(mipLevel);
		var pixelCount = mipData.Length;

		// Store the RGB values in an array format suitable for quantize
		// function

		// numRegardedPixels must be rounded up to avoid an
		// ArrayIndexOutOfBoundsException if all pixels are good.

		var numRegardedPixels = (quality <= 0) ? 0 : (pixelCount + quality - 1) / quality;

		var numUsedPixels = 0;
		var pixelArray = new int[numRegardedPixels][];

		for (var i = 0; i < pixelCount; i += quality) {
			var b = Convert.ToInt32(mipData[i].b);
			var g = Convert.ToInt32(mipData[i].g);
			var r = Convert.ToInt32(mipData[i].r);
			var a = Convert.ToInt32(mipData[i].a);

			// If pixel is mostly opaque and not white
			if (a >= 125 && !(ignoreWhite && r > 250 && g > 250 && b > 250)) {
				pixelArray[numUsedPixels] = new[] { r, g, b };
				numUsedPixels++;
			}
		}

		// Remove unused pixels from the array
		var copy = new int[numUsedPixels][];
		Array.Copy(pixelArray, copy, numUsedPixels);
		return copy;
	}

It could be further sped up if a higher mip level would be used and if the rest of the logic directly operated on the returned COLOR32 values, I guess.

Not sure if this is a generic solution for all cases, but for me, it worked wonders.