Reflection Glossiness not physically accurate?

[Felix2]

Hi,

As far as I've understood, a material with 100% white diffuse and no reflection should be identical to a material with 100% reflection and zero glossiness (if they're rendered with enough samples, GI, etc).


I made a simple scene to test this but apparently VRay can't reproduce diffuse reflectance with zero glossiness reflections as ball 5 should look somewhat identical to ball 6.
The above scene consists of a ground plane, a single VRayLight, the balls and a VRayPhysicalCam rendered with light cache for secondary GI and irradiance map for primary GI with the "High"-preset.

So what's up with this? Have I misunderstood something, done the test scene wrong or does the VRayMaterial have some sort of optimization which causes physically inaccurate results?

[vlado]

The reflections are physically correct for the implemented reflection models. However zero glossiness is not equal to pure diffuse reflection for the Phong, Blinn and Ward models (simply because of the way their authors defined them).

Best regards,
Vlado

[joconnell]

Yep = it might be that a white ball with no reflectance receives really ambient looking light - If you make a 100% reflective ball with no glossiness you're making a mirror. Try making a 100% reflective ball with 0 glossiness so that it's totally blurred out - should have a closer result to the white ball anyway. Gah - ye got there already

[dlparisi]

Well #5 (actually 1-5) are all refelcting the rest of the scene where #6 is not so I don't think they'll ever match. If you set the vraymtl to not trace reflections it'll generate just the highlight from the light and it matches much closer. Not exact, but pretty darn close.

[Morbid Angel]

interesting test, what gets to me is in ball 5 which is the glossiest of them all, the reflection uptop where the light source is the strongest in previous 1-4, is rather overtaken and is almost black. Without knowing the math behind this, to me visually that looks wrong.

[MasterBercon]

vlado, how hard would it be implmenting some more BRDF models like Cook–Torrance for example? And how heavy performance hit would it be to use actual measured BRDF data?

[vlado]

vlado, how hard would it be implmenting some more BRDF models like Cook–Torrance for example?

The Cook-Torrance model does not have efficient formulae for importance sampling, which makes implementation of glossy reflections slow.

And how heavy performance hit would it be to use actual measured BRDF data?

We have done that; it is not too slow, but has usability drawbacks:
(*) Additional storage is required, for each material;
(*) Mapping of the properties of such materials (e.g. glossiness) is not trivial;
(*) It is not trivial to actually measure the materials accurately;

Best regards,
Vlado

[^Lele^]

the problem with measured BRDF is always the fact that once that is it, that is it, there's no latitude to change it, at least in 3d.
And if you change it, then it's not like the original anymore...
Also, i don't think the issue with that stuff would be on the shading side, but rather on the amount of data a measured BRDF would be on disk and in memory.
Vlado may be able to be precise regarding this, but i heard from a bird, a while back, that there may be something nicer in terms of shading models, cooking...

[^Lele^]

good grief he's quick...