BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Australia/Brisbane
X-LIC-LOCATION:Australia/Brisbane
BEGIN:STANDARD
DTSTART:19920301T030000
TZOFFSETFROM:+1100
TZOFFSETTO:+1000
TZNAME:AEST
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20191203T080637Z
LOCATION:Great Hall 1&2
DTSTART;TZID=Australia/Brisbane:20191117T194000
DTEND;TZID=Australia/Brisbane:20191117T194100
UID:siggraphasia_SIGGRAPH Asia 2019_sess222_papers_190@linklings.com
SUMMARY:Reparameterizing Discontinuous Integrands for Differentiable Rende
ring
DESCRIPTION:Technical Papers Fast-Forward \n\nReparameterizing Discontinuo
us Integrands for Differentiable Rendering\n\nLoubet, Holzschuch, Jakob\n\
nDifferentiable rendering has recently opened the door to a number of chal
lenging inverse problems involving photorealistic images, such as computat
ional material design and scattering-aware reconstruction of geometry and
materials from photographs. Differentiable rendering algorithms strive to
estimate partial derivatives of pixels in a rendered image with respect to
scene parameters, which is difficult because visibility changes are inher
ently non-differentiable.\n\nWe propose a new technique for differentiatin
g path-traced images with respect to scene parameters that affect visibili
ty, including the position of cameras, light sources, and vertices in tria
ngle meshes. Our algorithm computes the gradients of illumination integral
s by applying changes of variables that remove or strongly reduce the depe
ndence of the position of discontinuities on differentiable scene paramete
rs. The underlying parameterization is created on the fly for each integra
l and enables accurate gradient estimates using standard Monte Carlo sampl
ing in conjunction with automatic differentiation. Importantly, our approa
ch does not rely on sampling silhouette edges, which has been a bottleneck
in previous work and tends to produce high-variance gradients when import
ant edges are found with insufficient probability in scenes with complex v
isibility and high-resolution geometry. We show that our method only requi
res a few samples to produce gradients with low bias and variance for chal
lenging cases such as glossy reflections and shadows. Finally, we use our
differentiable path tracer to reconstruct the 3D geometry and materials of
several real-world objects from a set of reference photographs.\n\nRegist
ration Category: Platinum Pass, Full Conference Pass, Full Conference One-
Day Pass, Basic Conference Pass, Student One-Day Pass, Exhibitor Pass
URL:https://sa2019.conference-program.com/presentation?id=papers_190&sess=
sess222
END:VEVENT
END:VCALENDAR