Colorimetrically clip after gamut mapping
The simplest way to do this is probably by just linearly desaturating towards vec3(luma) until all three components are within bounds.
Questions:
- Do 'out of gamut' colors ever contain components above 1, or is just negative values that we need to worry about?
- How hard would it be to do something smarter like L*Ch? In what ways does scaling C differentiate from mixing with
vec3(luma)? - How to detect whether one triangle lies fully inside another? (my guess is something like "check whether all three points of the inner tri are on the same side of all three sides of the outer tri")
We could also do something smarter like soft-knee in the future, which might require somehow bringing our desaturation coefficient into a parameter space where [0-1] represent in-gamut values and values above 1 represent "out of gamut".
Activity
added shaders label
Do 'out of gamut' colors ever contain components above 1, or is just negative values that we need to worry about?
The answer is 'yes', but that might still be the same thing. I suspect that all out of gamut values will have negative components, and that by reducing the saturation until all components are nonnegative, we also automatically constrain the largest component to be below 1.
Interesting questions need to be explored w.r.t how tone mapping and gamut mapping interact, i.e. whether tone mapping (mix towards black) commutes with gamut mapping (mix towards luma) and in which order we want to be performing operations.
I suspect that all out of gamut values will have negative components, and that by reducing the saturation until all components are nonnegative, we also automatically constrain the largest component to be below 1.
This is also not true. Some out-of-gamut colors will end up simply "too bright", similar to HDR colors. Fixed this by adding a second pass to scale the color down in brightness (linearly) until it's in range again.
Edited by Niklas Haasclosed via commit d63eeb1e
mentioned in commit d63eeb1e
