Add complaints

waacton · May 9, 2024 · 91f3ab8 · 91f3ab8
1 parent 63f1fb6
commit 91f3ab8
Show file tree

Hide file tree

Showing 2 changed files with 12 additions and 2 deletions.
diff --git a/Unicolour/Oklab.cs b/Unicolour/Oklab.cs
@@ -53,6 +53,11 @@ internal Oklab(double l, double a, double b, ColourHeritage heritage) : base(l,
      * LMS = M1 * RgbToXyz * Rgb
      * RgbToOklab = M1 * RgbToXyz
      * M1 = RgbToOklab * RgbToXyz^-1
+     * --------------------
+     * 💩💩💩 the "proper" way to define M1 is hidden in some random GitHub thread, instead of in any Oklab documentation
+     * https://github.com/w3c/csswg-drafts/issues/6642#issuecomment-945714988
+     * modern colour spaces seem to make it hard to implement things accurately (see also: XYB)
+     * will consider reworking M1 calculation if there's ever any complaints to my reverse engineering
      */
 
     private static readonly WhitePoint OklabWhitePoint = Illuminant.D65.GetWhitePoint(Observer.Degree2);

diff --git a/Unicolour/Xyb.cs b/Unicolour/Xyb.cs
@@ -20,7 +20,7 @@ internal Xyb(double x, double y, double b, ColourHeritage heritage) : base(x, y,
     public override string ToString() => base.ToString();
 
     /*
-     * XYB is a transform of RGB
+     * XYB is a transform of RGB Linear
      * Forward: https://ds.jpeg.org/whitepapers/jpeg-xl-whitepaper.pdf
      * Reverse: n/a - not provided, using own implementation
      * 
@@ -32,7 +32,12 @@ internal Xyb(double x, double y, double b, ColourHeritage heritage) : base(x, y,
     /*
      * NOTE: the final step of B -= Y is not documented in the JPEG XL white paper
      * but apparently the intention is that B = sGamma should be B = sGamma - Y 🤷
-     * (at least it makes the colour greyscale when X == 0 and B == 0, in the same way as LAB)
+     * (at least it makes the colour greyscale when X == 0 and B == 0, in the same way as LAB-like spaces)
+     * --------------------
+     * 💩💩💩 the only mention of this "subtract Y from B" is just some random Twitter post and a 10 second YouTube video
+     * so it's not particularly convincing that this extra step, which is left out of the white paper, should be performed
+     * https://github.com/Evercoder/culori/issues/200 · https://twitter.com/jonsneyers/status/1605321352143331328 · https://www.youtube.com/watch?v=rvhf6feXw7w
+     * modern colour spaces seem to make it hard to implement things accurately (see also: Oklab)
      */
 
     private const double Bias = 0.00379307325527544933;