BRDF Materials

Per-Triangle

Key problem: Whatever BRDF we have, we need to sample it (generate random rays with probability given by BRDF).

Brute Force method: Generate random rays (uniform), calculate BRDF to attenuate color. -> Bad convergence to expect.

Opposed to: Importance Sampling. Sample either with exact probability or with almost correct probability and weight difference. There is an analytical solution for phong! According to multiple other sources, there are no analytical solutions to other BRDFs. However, this isn't necessarily the best method.. (may resemble paths that are important to not very well)

Resources on the topic:

• (The Brigade Engine Guy's Master Thesis (Must be good ;))) http://repository.tudelft.nl/view/ir/uuid%3A4a5be464-dc52-4bd0-9ede-faefdaff8be6/
  • Good explanation of Importance Sampling on page 25, exactly what I searched for
• (2004 paper) Efficient BRDF Importance Sampling Using a Factored Representation http://cseweb.ucsd.edu/~ravir/papers/brdf/
  • This is one of the candidates for the hierarchy. They use SVD to depict arbitrary BRDFs.
• (2010 course) Importance Sampling for Production Rendering http://www.igorsklyar.com/system/documents/papers/4/fiscourse.comp.pdf
  • Derives analytical phong (comparatively easy)
  • Some infos to good Cook-Torrance sampling
  • Much infos about sampling of env probes
  • Complicated and lengthy
• (GPUGems3) http://http.developer.nvidia.com/GPUGems3/gpugems3_ch20.html
  • Phong again yawn
  • AaaaHA! "Similar importance sampling methods can be used for other material models, such as the Lafortune BRDF (Lafortune et al. 1997) or Ward's anisotropic BRDF (Walter 2005). Pharr and Humphreys 2004 gives more details on deriving sampling formulas for various BRDFs." This refers to the book "Physically Based Rendering: From Theory to Implementation"
• http://digibug.ugr.es/bitstream/10481/19751/1/rmontes_LSI-2012-001TR.pdf
  • Table on page 4 denotes with a star if a BRDF has an importance sampling strategy
  • Section 5: explains sampling for: Ideal Diffuse, Ideal Specular, Lobes (Phong, Lafortune, ..), HalfAngle based Lobes (Blinn, ..)
  • General methods rely on wavelet or similar -> require more space for precomputed stuff (200KB up to multiple MB)
• Generic BRDF Sampling
  • The method where the pictures in the survey come from
  • requres ~90 Quadtrees per BRDF(parametrization) -> 1MB mem.

General BRDF Notes

Talk Slides of a good experimantal comparision of basic BRDFs.

Hierarchical BRDF

TODO