> > I do expect benefits in all areas that include colored distance-based
> > attenuation, such as with fog, absorbing media (including the extinction
> > effect of scattering media) and interior fading.
>
> Won't it also vastly improve any scenes where the "light" is not
> broadband, eg street lights, LEDs, lasers etc. Would fluorescent
> materials be supported?

Yes, it will give more physically correct render for strange lights. But the
most important aspect of spectral rendering is more "natural" color behaviour.
The problem of rgb illumination model is that the primary colors are
"orthogonal". So when you say "blue", and "red" they have absolutely nothing in
common. So a red light will give no illumination to red object, and red and blue
objects will have no radiosity effect on each other at all. This is a bit
unexpected usually - you put a primary colored object in a scene, and it seems
almost monochromatic, even if you have reflections and colored lights in a
system - it does not "take" other colors. So in essence, radiosity has too
strong contrast, colors are too sharp, and primary colors don't mix well.

I usually "fake" color overlap by having colors such as <1,0.1,0.1> for red and
so on. But it's not the same. Only spectral rendering will give you true
richness of color - you can have a pure intense color, but it can still react to
environmental illumination in a correct way. A typical example is a car in a
street.

Spectral rendering gives you subtle differences that you don't notice, but you
do notice it looks better. Just like focal blur, HDR environment mapping and
difference between flat ambient and radiosity.

> In theory colometrically correct renders could be produced which I
> imagine would open up new markets for POV use (especially if combined
> with an unbiased tracing option). We need something to keep increasing
> those render times to keep up with new hardware :-)

Sure :) But in this case, I don't even think it will affect the render time that
much. Most of the rendering cost is the raytracing (finding intersections and
normals). Color computation is only a small part of the process, so unless you
have color dispersion in refractions and reflections, you will get a small
overhead (especially if the C++ core behind it is written efficiently).


About flourescence: it seems like a very nice idea. But somehow this opens a
wide array of options (every spectral input color could potentially produce a
different output color of different intensity). It smells more like a
custom-made shader. Having a pigment that is actually a function of input light
color, direction and coordinates, would make more sense in this case, and would
allow flexibility for users to design funny materials (like angle-dependent
colors, fluorescence, anisotropic BDRF functions, color-dependent specularity
angle and possibly a nonlinear response). This is quite a lot to ask from
developers, but it would get easier, if LLVM support actually happens at some
point in the future.

Post a reply to this message