BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Australia/Melbourne
X-LIC-LOCATION:Australia/Melbourne
BEGIN:DAYLIGHT
TZOFFSETFROM:+1000
TZOFFSETTO:+1100
TZNAME:AEDT
DTSTART:19721003T020000
RRULE:FREQ=YEARLY;BYMONTH=4;BYDAY=1SU
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:19721003T020000
TZOFFSETFROM:+1100
TZOFFSETTO:+1000
TZNAME:AEST
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20240214T070245Z
LOCATION:Meeting Room C4.11\, Level 4 (Convention Centre)
DTSTART;TZID=Australia/Melbourne:20231214T090000
DTEND;TZID=Australia/Melbourne:20231214T091500
UID:siggraphasia_SIGGRAPH Asia 2023_sess124_papers_891@linklings.com
SUMMARY:BakedAvatar: Baking Neural Fields for Real-Time Head Avatar Synthe
 sis
DESCRIPTION:Technical Papers, TOG\n\nHao-Bin Duan (Beihang University); Mi
 ao Wang (Beihang University, Zhongguancun Laboratory); Jin-Chuan Shi and X
 u-Chuan Chen (Beihang University); and Yan-Pei Cao (Tencent)\n\nSynthesizi
 ng photorealistic 4D human head avatars from videos is essential for VR/AR
 , telepresence, and video game applications. Although existing Neural Radi
 ance Fields (NeRF)-based methods achieve high-fidelity results, the comput
 ational expense limits their use in real-time applications. To overcome th
 is limitation, we introduce BakedAvatar, a novel representation for real-t
 ime neural head avatar synthesis, deployable in a standard polygon rasteri
 zation pipeline. Our approach extracts deformable multi-layer meshes from 
 learned isosurfaces of the head and computes expression-, pose-, and view-
 dependent appearances that can be baked into static textures for efficient
  rasterization. We thus propose a three-stage pipeline for neural head ava
 tar synthesis, which includes learning continuous deformation, manifold, a
 nd radiance fields, extracting layered meshes and textures, and fine-tunin
 g texture details with differential rasterization. Experimental results de
 monstrate that our representation generates synthesis results of comparabl
 e quality to other state-of-the-art methods while significantly reducing t
 he inference time required. We further showcase various head avatar synthe
 sis results from monocular videos, including view synthesis, face reenactm
 ent, expression editing, and pose editing, all at interactive frame rates.
 \n\nRegistration Category: Full Access\n\nSession Chair: Lin Gao (Universi
 ty of Chinese Academy of Sciences)
URL:https://asia.siggraph.org/2023/full-program?id=papers_891&sess=sess124
END:VEVENT
END:VCALENDAR