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:20260817T171531Z
LOCATION:Hall B5 (1)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241205T093400
DTEND;TZID=Asia/Tokyo:20241205T094600
UID:siggraphasia_SIGGRAPH Asia 2024_sess124_papers_852@linklings.com
SUMMARY:Deformation Recovery: Localized Learning for Detail-Preserving Def
 ormations
DESCRIPTION:Ramana Sundararaman (Centre National de la Recherche Scientifi
 que - Laboratoire d'informatique de l'École Polytechnique (LIX)); Nicolas 
 Donati (Ansys); Simone Melzi (University of Milano-Bicocca); Etienne Corma
 n (Université de Lorraine, CNRS); and Maks Ovsjanikov (Centre National de 
 la Recherche Scientifique - Laboratoire d'informatique de l'École Polytech
 nique (LIX))\n\nWe introduce a novel data-driven approach aimed at designi
 ng high-quality shape deformations based on a coarse localized input signa
 l. Unlike previous data-driven methods that require a global shape encodin
 g, we observe that detail-preserving deformations can be estimated reliabl
 y without any global context in certain scenarios. Building on this intuit
 ion, we leverage Jacobians defined in a one-ring neighborhood as a coarse 
 representation of the deformation. Using this as the input to our neural n
 etwork, we apply a series of MLPs combined with feature smoothing to learn
  the Jacobian corresponding to the detail-preserving deformation, from whi
 ch the embedding is recovered by the standard Poisson solve. Crucially, by
  removing the dependence on a global encoding, every point becomes a train
 ing example, making the supervision particularly lightweight. Moreover, wh
 en trained on a class of shapes, our approach demonstrates remarkable gene
 ralization across different object categories. Equipped with this novel ne
 twork, we explore three main tasks: refining an approximate shape correspo
 ndence, unsupervised deformation and mapping, and shape editing.\n\nRegist
 ration Category: Full Access, Full Access Supporter\n\nLanguage Format: En
 glish Language\n\nSession Chair: Yotam Gingold (George Mason University)\n
 \n
URL:https://asia.siggraph.org/2024/program/?id=papers_852&sess=sess124
END:VEVENT
END:VCALENDAR