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:20260817T171530Z
LOCATION:Hall B5 (1)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241206T110800
DTEND;TZID=Asia/Tokyo:20241206T111900
UID:siggraphasia_SIGGRAPH Asia 2024_sess142_papers_220@linklings.com
SUMMARY:Reconstruct translucent thin objects from photos
DESCRIPTION:Xi Deng (Cornell University); Lifan Wu (NVIDIA Research); Bruc
 e Walter (Cornell University); Ravi Ramamoorthi (University of California 
 San Diego, NVIDIA Research); Eugene d'Eon (NVIDIA Research); Steve Marschn
 er (Cornell University, NVIDIA Research); and Andrea Weidlich (NVIDIA Rese
 arch)\n\nThe joint reconstruction of shape and appearance for translucent 
 objects from real-world data poses a challenge in computer graphics, espec
 ially when dealing with complex layered materials like leaves or paper. Th
 e traditional assumption of diffuse transmittance falls short, and more ac
 curate Monte-Carlo-based models are often needed to reproduce their appear
 ance. To accurately capture the translucent appearance, an acquisition sys
 tem needs to be carefully designed. Additionally, there are three challeng
 es for inverse rendering: First, a large number of unknown parameters make
  the optimization problem difficult. Second, the Monte Carlo (MC) renderer
  introduces noise, which the optimization is sensitive to, \nespecially wh
 en dealing with complex material models such as rough dielectric surfaces 
 and highly scattering participating media. Last, MC estimators using long 
 light paths (more than 32 bounces in our case) create a large computation 
 graph in memory,  making the gradient back-propagation costly.\nTo address
  those challenges, we present a cheap and fast acquisition pipeline that c
 an capture spatially-varying reflectance and transmission at the same time
 , using a two-phase optimization. We first initialize the geometry with th
 e traditional vision method and then fit a simple and fast appearance mode
 l. Thereafter, we use the estimated parameters to initialize a second opti
 mization using a more expensive volumetric model, which converges faster a
 nd more reliably from this favorable starting position.  We also introduce
  a way to analyze each parameter's sensitivity to the noise in the measure
 ments, which can be used in optimally selecting useful measurements for op
 timization. Furthermore, instead of iterating on the camera system, we als
 o introduce an optimal weighted $\ell_2$ loss as an alternative for select
 ing useful pixels from existing measurements.\n\nRegistration Category: Fu
 ll Access, Full Access Supporter\n\nLanguage Format: English Language\n\nS
 ession Chair: Maria Larsson (The University of Tokyo)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_220&sess=sess142
END:VEVENT
END:VCALENDAR