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:20250110T023313Z
LOCATION:Hall B5 (2)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241206T092300
DTEND;TZID=Asia/Tokyo:20241206T093400
UID:siggraphasia_SIGGRAPH Asia 2024_sess140_papers_281@linklings.com
SUMMARY:Neural Differential Appearance Equations
DESCRIPTION:Technical Papers\n\nChen Liu and Tobias Ritschel (University C
 ollege London (UCL))\n\nWe propose a method to reproduce dynamic appearanc
 e textures with space-stationary but time-varying visual statistics.\nWhil
 e most previous work decomposes dynamic textures into static appearance an
 d motion, we focus on dynamic appearance that results not from motion but 
 variations of fundamental properties, such as rusting, decaying, melting, 
 and weathering.\nTo this end, we adopt the neural ordinary differential eq
 uation (ODE) to learn the underlying dynamics of appearance from a target 
 exemplar.\nWe simulate the ODE in two phases.\nAt the ``warm-up'' phase, t
 he ODE diffuses a random noise to an initial state.\nWe then constrain the
  further evolution of this ODE to replicate the evolution of visual featur
 e statistics in the exemplar during the generation phase.\nThe particular 
 innovation of this work is the neural ODE achieving both denoising and evo
 lution for dynamics synthesis, with a proposed temporal training scheme.\n
 We study both relightable (BRDF) and non-relightable (RGB) appearance mode
 ls.\nFor both we introduce new pilot datasets, allowing, for the first tim
 e, to study such phenomena:\nFor RGB we provide 22 dynamic textures acquir
 ed from free online sources;\nFor BRDF, we further acquire a dataset of 21
  flash-lit videos of time-varying materials, enabled by a simple-to-constr
 uct setup.\nOur experiments show that our method consistently yields reali
 stic and coherent results, whereas prior works falter under pronounced tem
 poral appearance variations.\nA user study confirms our approach is prefer
 red to previous work for such exemplars.\n\nRegistration Category: Full Ac
 cess, Full Access Supporter\n\nLanguage Format: English Language\n\nSessio
 n Chair: Seungyong Lee (POSTECH)
URL:https://asia.siggraph.org/2024/program/?id=papers_281&sess=sess140
END:VEVENT
END:VCALENDAR