#### Project Overview:

Accurate representations of real-world materials are a crucial prerequisite for realistic image synthesis. This project focuses on modeling cloth appearance based on light scattering from individual threads. Making use of reflectance acquisition techniques, we have studied a variety of threads extracted from real-world fabrics.

Based on our reflectance data, we’ve come up with a flexible physically based approach that is able to represent the reflectance function of (theoretically) all fabrics. The detailed study and reflectance measurement of fabrics is described in my thesis and a slightly different flavor in the follow-up Siggraph paper. There is also a discussion and explanation of the implementation details. Future may involve an importance-sampled version of the shader or a real-time approximation version.

#### Publications:

A Measurement-Based Appearance Model for Fabrics
Oleg Bisker
Thesis (UCSD 2011)

#### Discussion:

Sampling

The cloth BRDF sampling formulation described in the Siggraph paper is the vanilla flavor of a multitude of possible approaches. It uniformly picks samples from each of the orthogonal tangent curves and computes a weighted average based on relative visibility. In practice, this can be quite costly especially for curves with few flat sections.

An alternative sampling scheme is to focus the samples based on where the reweighting function has the highest values. In an MC-style renderer this gives us an instant preview and reduced workload by not wasting samples on portions of the tangent curve which are self-occluded or minimally contributing. The code implements a very basic version of this concept to reduce excess thread BSDF evaluation. In the future, we plan to extend the importance sampling to take into account the shadowing/masking and possibly the actual scattering lobes.

Implementation Details

Accurately computing the angles in the right ranges is step one. By projecting the eye/light vectors into a tangent-space we can easily compute the three angle quantities used in the paper: $$\theta, \phi, \psi$$. There are alternative derivations for the angles that avoid transcendental functions, however they are less concise and easier to get wrong.
 $$$$\quad a = dot(\omega, t')\\ \quad b = dot(\omega, n)\\ \quad c = dot(\omega, t)\\$$$$ $$$$\quad \theta = asin(c)\\ \quad \phi = atan(a, b)\\ \quad \psi = atan(c, b)\\$$$$

Relevant Code - [View Shader Source] -

Params: tangent basis vectors
Returns: tangent frame Vec3f($$\theta, \phi, \psi$$) angles used in BSDF:
Vec3f getTangentBasisAngles(n,t, t’, w)

Scattering contribution from a single thread sample.
Params: tangent basis angles and thread attributes
Returns: thread scattering Vec2f(volume scatter, surface reflect):
Vec2f threadBSDF($$\theta_i, \phi_i, \theta_r, \phi_r$$, lobeWidth, ior, kd)

Vec2f weaveBRDFSample(n, t, t’, $$\omega_i, \omega_o$$, thread, ...)
Col3f weaveBRDF(n, t, t’, $$\omega_i, \omega_o$$)