BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Asia/Tokyo
X-LIC-LOCATION:Asia/Tokyo
BEGIN:STANDARD
TZOFFSETFROM:+0900
TZOFFSETTO:+0900
TZNAME:JST
DTSTART:18871231T000000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20260817T171534Z
LOCATION:Hall B5 (2)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241206T094600
DTEND;TZID=Asia/Tokyo:20241206T095800
UID:siggraphasia_SIGGRAPH Asia 2024_sess140_papers_551@linklings.com
SUMMARY:Markov-Chain Monte Carlo Sampling of Visibility Boundaries for Dif
 ferentiable Rendering
DESCRIPTION:Peiyu Xu (University of California Irvine); Sai Bangaru (MIT C
 SAIL, NVIDIA Research); Tzu-Mao Li (University of California San Diego); a
 nd Shuang Zhao (University of California Irvine)\n\nPhysics-based differen
 tiable rendering requires estimating boundary path integrals emerging from
  the shift of discontinuities (e.g., visibility boundaries). Previously, a
 lthough the mathematical formulation of boundary path integrals has been e
 stablished, efficient and robust estimation of these integrals has remaine
 d challenging. Specifically, state-of-the-art boundary sampling methods al
 l rely on primary-sample-space guiding precomputed using sophisticated dat
 a structures---whose performance tends to degrade for finely tessellated g
 eometries.\n\nIn this paper, we address this problem by introducing} a new
  Markov-Chain-Monte-Carlo (MCMC) method. At the core of our technique is a
  local perturbation step capable of efficiently exploring highly fragmente
 d primary sample spaces via specifically designed jumping rules.\nWe compa
 re the performance of our technique with several state-of-the-art baseline
 s using synthetic differentiable-rendering and inverse-rendering experimen
 ts.\n\nRegistration Category: Full Access, Full Access Supporter\n\nLangua
 ge Format: English Language\n\nSession Chair: Seungyong Lee (POSTECH)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_551&sess=sess140
END:VEVENT
END:VCALENDAR