Commerce tone mapper

[elalish]

This adds an experimental feature, as we're still soliciting feedback on the subtleties of the equation before it will become an official part of our API. The new feature is a tone mapping function designed for 3D Commerce applications as an alternative to our default, ACES. The Commerce tone mapper is designed to retain color accuracy as much as possible without producing noticeable artifacts in highlights (ACES and Linear/Clamped/No tone mapping both have poor color accuracy, though in different areas).

Note: the Commerce option has less contrast than ACES (intentionally), which may make it look "flat" in a side-by-side comparison with ACES. In general, higher contrast lighting should be used with Commerce tone mapping than with ACES to make up the difference. We may add a related option for our neutral lighting in the future, since the current default was designed for ACES.

This PR also adds an option for testing this in our Editor, as well as an additional model, Macbeth, that helps to show the color issues we're attempting to resolve. Macbeth includes unlit (and therefore un-tone-mapped) spheres with the same base color as the shiny and matte spheres to allow comparison.

ACES:

Commerce:

[hybridherbst]

I'm seeing a color shift on higher exposures; e.g. this model starts to become red-ish at exposure = 3 or higher: (the editor only allows up to 2 right now) van-veer-original-custom.zip	Exposure=1	Exposure=3	Exposure=10

Is this intended?

And an additional question: why is peak calculated with max(r, max(g,b)) instead of using a perceptual luminance value like float peak = dot(color, vec3(0.299, 0.587, 0.114)) or so?

[elalish]

Thanks for the feedback! This appears intentional to me, since brown is also red-ish, just darker. With white light, PBR will only multiply your base color by a scalar factor and add some white - this tone mapper does the same: multiplies the color by a scalar factor and mixes toward white (desaturates for high brightness). Do you have any examples of what you would expect this color to look like when overexposed?

[elalish]

As to perceptual luminance, I tried that route first and found it actually introduced a lot of problems. I finally realized it was because the purpose of a tone mapper is to fit into the sRGB unit cube, whose bounds do not have uniform perceptual luminance. That vector is useful when working in a luminance space, but I am not.