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:20260817T171516Z
LOCATION:Hall C\, C Block\, Level 4
DTSTART;TZID=Asia/Tokyo:20241203T090000
DTEND;TZID=Asia/Tokyo:20241203T120000
UID:siggraphasia_SIGGRAPH Asia 2024_sess223_papers_805@linklings.com
SUMMARY:Manifold Sampling for Differentiable Uncertainty in Radiance Field
 s
DESCRIPTION:Linjie Lyu (Max Planck Institute for Informatics), Ayush Tewar
 i (MIT CSAIL), Marc Habermann (Max Planck Institute for Informatics), Shun
 suke Saito and Michael Zollhöfer (Meta Codec Avatars Lab), and Thomas Leim
 kühler and Christian Theobalt (Max Planck Institute for Informatics)\n\nRa
 diance fields are powerful and, hence, popular models for representing the
  appearance of complex scenes. Yet, constructing them based on image obser
 vations gives rise to ambiguities and uncertainties. We propose a versatil
 e approach for learning Gaussian radiance fields with explicit and fine-gr
 ained uncertainty estimates that impose only little additional cost compar
 ed to uncertainty-agnostic training. Our key observation is that uncertain
 ties can be modeled as a low-dimensional manifold in the space of radiance
  field parameters that is highly amenable to Monte Carlo sampling. Importa
 ntly, our uncertainties are differentiable and, thus, allow for gradient-b
 ased optimization of subsequent captures that optimally reduce ambiguities
 . We demonstrate state-of-the-art performance on next-best-view planning t
 asks, including high-dimensional illumination planning for optimal radianc
 e field relighting quality.\n\nRegistration Category: Enhanced Access, Ful
 l Access, Full Access Supporter\n\nLanguage Format: English Language\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_805&sess=sess131
END:VEVENT
END:VCALENDAR