(L) [2013/11/26] [tby ypoissant] [Heckbert ray grammar and today BRDF] Wayback!I'm wondering how do you classify a ray that originates from a smooth BRDF? A BRDF that is smooth but is not perfect Mirror?
The original Heckbert paper "Adaptive Rasiosity Textures for Bidirectional Ray Tracing" was published in 1990 and the 'S' classification was specifically for perfect Mirror. See his simplifying assumption #6: "Specular surfaces are not rough. All specular interaction is ideal." where his definition of "rough specular" is a smooth BRDF. And of course the 'D' classification if for perfect Lambertian surfaces.
I see Heckbert grammar still used in relatively recent articles and thesis about MLT and derivatives in particular because mutation strategies are based on whether a surface is specular or not. Is it because such MLT derived systems still assume that surfaces are either perfect diffuse or perfect smooth? If not how would a smooth BRDF be classified then?
(L) [2013/11/26] [tby Dietger] [Heckbert ray grammar and today BRDF] Wayback!The research community has secretly agreed to ignore glossy materials, too complicated. Now that you are in on this secret.... don't tell anyone! [SMILEY :D]
Seriously though; In the context of MLT 'D' usually just stands for everything-that-is-not-'S'. Some people use a special symbol such as 'G' to identify glossy surfaces, but this is kind of a work-around. The issue is of course that materials cannot really be classified in two (or three) groups, glossy materials can be arbitrarily close to perfect diffuse or perfect specular. Assuming you adjust the pdf computations appropriately, MLT mutations usually still works if you replace diffuse bounces by not-too-glossy bounces. However, mutation strategies may become very ineffective when replacing diffuse bounces by highly glossy bounces. Similarly, replacing specular bounces by very-glossy bounces may also be ok on some occasions. Mixing the two approaches as done done by Wenzel Jacob for Manifold Exploration should work reasonably well because usually at least one mutation strategy will be effective.