UCSD CSE 272 : Advanced Image Synthesis --- advanced BSDF

Layered BSDF

how to modeling layered BSDF

adding equation: 把每一层的光照作用累积起来

use a Fourier basis to compress the matrices https://www.pbr-book.org/3ed-2018/Reflection_Models/Fourier_Basis_BSDFs
downside
- need to store a huge sparse matrix (in Fourier domain)
- texturing is hard

just compute the multiple scattering using Monte Carlo integration
降维
- the horizontal distance doesn’t matter in a BSDF
- solve a small “1D” rendering equation inside the material

option1 --- polylines(多线段) but lines don’t have thickness!
option 2 --- connected cylinders(圆柱) rendering cylinders is a bit expensive
option 3 --- connected quads(有点类似于billboard的四边形)
- quads always face towards the viewers
- use shading normal to fake cylinder appearances
option 4 --- connected Bezier curves
- smoother appearance and less storage cost
- can use either the planar approximation or have actual cylindrical shape
- can use shading normal to represent orientation variation within the curve

or “BCSDF” (Bidirectional Curve Scattering Distribution Function) or “BFSDF” (F=Fiber)

lighting model
- use a lobe for each scattering mode
- Common assumption: separable lobes for longitudinal & azimuthal
longitudinal scattering
- Marschner et al. [2003]: fit different wrapped Gaussians around the half-angle
- d’Eon et al. [2011]’s proposal: use a von-Mises-like distribution http://www.eugenedeon.com/wp-content/uploads/2014/04/egsr
azimuthal scattering Common assumption: the circular section is isotropic (i.e. it’s not an ellipse)
- Chiang et al. [2016] proposed to use a logistic distribution
volumetric absorption and Fresnel 3 lobes for R, TRT, TT; 1 lobe to compensate energy
Bells and whistles(really interesting)
- 动物毛发和人类毛发的差别: volumetric scattering
- 不同人种真实的头发截面形状: elliptical cross section
- 不区分longitudinal和azimuthal: non-separable lobes
- real-time rendering