i’ve been looking at hue constant colour spaces again (and again) for ui interaction this time. you would think that oklab is a lot better than using completely nonsensical spaces like say HSV or CIELab. in fact, this is what it looks like if you plot same hue over different value levels (y coordinate in the plot):
i’m trying to select the yellow flower here in the center… needless to say the oklab visualisation is a bit useless here. the values are tonemapped for display so the gradient is not same smooth everywhere.
sigh i think even for simple ui visualisation there is no way around walking munsell data
Hi,
Thanks for the post, but I’m not sure I understand what I’m looking at. Can you maybe show the formulas corresponding to the plots, or some kind of pseudo code?
ah… it’s not very deep… i was just plotting lightness vs. hue as y vs. x for instance in the mask plot in the first image, for the OKLab colour space. i was shocked to see how much the hue wiggles when you change L. an ideal UCS wouldn’t twitch with y.
this is kinda expected because as far as i understand OKLab was derived to give nice gradients for same hue with varying colourfulness (sliced the other dimension).
but the application here is of course masking: select a hue in the plot and see how it masks out the yellow flower in this case (which you can’t see because it’s black now). this is rendered completely useless if you show the hue at some lightness that will render as red instead of yellow.
just in very slow, walking a triangulation of the data.
