Once-more scattered next event estimation

Johannes Hanika, Andrea Weidlich, Marc Droske
Weta Digital,
Karlsruhe Institute of Technology, NVIDIA, Unity

© Mark Yokoyama (flickr)

© Phil's 1stPix (flickr)

© Charles EYES PiX (flickr)

© Lenka Reznicek (flickr)

© melfoody (flickr)

How different are the two effects?

for moderately forward scattering phase function

  1. Blue line: unit test collected histogram of contributions over all $\boldsymbol{x}_1$,
    plotted over outgoing angle, contains $L_i \cdot f_s$
  1. Actually the same effect (product of light and phase function)
  2. We know how to sample it: via joint importance sampling/tabulation

    [GKH*13] Georgiev et al., “Joint importance sampling of low-order volumetric scattering”, SIGGRAPH Asia 2013.

  3. Or evaluate analytically via series expansion of the phase function

    [PSP10] Pegoraro et al., “A closed-form solution to single scattering for general phase functions and light distributions”, EGSR 2010.

$g=0.98$

$g=0.5$

Circumcircle geometry

  1. "Waterlevel" in the demo determines $\theta$ at center vertex of triangle
  2. Picking any $\boldsymbol{x}_1$ on the arc (above waterlevel) has constant $\theta$!