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 (2)\, B Block\, Level 5
DTSTART;TZID=Asia/Tokyo:20241206T105600
DTEND;TZID=Asia/Tokyo:20241206T110800
UID:siggraphasia_SIGGRAPH Asia 2024_sess143_papers_468@linklings.com
SUMMARY:GaussianHeads: End-to-End Learning of Drivable Gaussian Head Avata
 rs from Coarse-to-fine Representations
DESCRIPTION:Kartik Teotia (Max Planck Institute for Informatics, Saarland 
 Informatics Campus); Hyeongwoo Kim (Imperial College London); Pablo Garrid
 o (Flawless AI); Marc Habermann (Max Planck Institute for Informatics, Saa
 rland Informatics Campus); Mohamed Elgharib (Max Planck Institute for Info
 rmatics); and Christian Theobalt (Max Planck Institute for Informatics, Sa
 arland Informatics Campus)\n\nReal-time rendering of human head avatars is
  a cornerstone of many computer graphics applications, such as augmented r
 eality, video games, and films, to name a few. Recent approaches address t
 his challenge with computationally efficient geometry primitives in a care
 fully calibrated multi-view setup. Albeit producing photorealistic head re
 nderings, it often fails to represent complex motion changes such as the m
 outh interior and strongly varying head poses. We propose a new method to 
 generate highly dynamic and deformable human head avatars from multi-view 
 imagery in real-time. At the core of our method is a hierarchical represen
 tation of head models that allows to capture the complex dynamics of facia
 l expressions and head movements. First, with rich facial features extract
 ed from raw input frames, we learn to deform the coarse facial geometry of
  the template mesh. We then initialize 3D Gaussians on the deformed surfac
 e and refine their positions in a fine step. We train this coarse-to-fine 
 facial avatar model along with the head pose as a learnable parameter in a
 n end-to-end framework. This enables not only controllable facial animatio
 n via video inputs, but also high-fidelity novel view synthesis of challen
 ging facial expressions, such as tongue deformations and fine-grained teet
 h structure under large motion changes. Moreover, it encourages the learne
 d head avatar to generalize towards new facial expressions and head poses 
 at inference time. We demonstrate the performance of our method with compa
 risons against the related methods on different datasets, spanning challen
 ging facial expression sequences across multiple identities. We also show 
 the potential application of our approach by demonstrating a cross-identit
 y facial performance transfer application.\n\nRegistration Category: Full 
 Access, Full Access Supporter\n\nLanguage Format: English Language\n\nSess
 ion Chair: Iain Matthews (Epic Games, Carnegie Mellon University)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_468&sess=sess143
END:VEVENT
END:VCALENDAR