The Multiple Scattering Microfacet BSDFs with the Smith Model paper describes a statistical model for replacing the masking-shadowing functions in microfacet BSDFs (which account for paths with more than one surface intersection by setting their contribution to 0) with a distribution which can be path traced and allows the a ray to intersect a microfacet surface several times before exiting.

They do this by modifying a volumetric (microflake) model to behave like a heightfield: to never collide with anything "above" the surface and to always collide with anything "below" the surface.

Microsurface vs modified microflake volume

The result, using wording from their slides, is that "the ray can never go through the Smith volume; the model creates an opaque surface-like interface."

Is this model compatible with traditional diffuse subsurface scattering, where a path can travel a perceptible (macro-scale) distance through a surface before exiting? Or is this simply a specular BSDF with no intention of modeling long paths internal to a surface, to be combined with a diffuse BSDF that adds that component?

(Does the microflake model of volumetric rendering normally separate diffuse and specular, or is "diffuse" simply a path that bounces around so many times that the outgoing direction is uniformly distributed?)


1 Answer 1


The goal of Heitz et al.'s model is pretty much the opposite of subsurface scattering: They only consider surface scattering, i.e. the ray can never enter the material.

Because microfacets are statistical in nature, they can recast their problem in such a way that it can be solved by microflakes, which allows them to compute properties such as the mean free path to derive a heightfield sampling procedure. However, even though they use microflake theory, they still solve the same problem of multiple scattering on microfacets, and their result is still a BSDF, not a subsurface scattering model.

Because it is a BSDF, the path exits at the same location as it hit the surface. This is also mentioned in the introduction:

Note that our volume-scattering process models the interaction with a surface, but its application in rendering is virtual in that no displacements occur, i.e. the incident and exitant location are the same and the resulting plane-parallel radiometry produces a BSDF.


Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.