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 B7 (1)\, B Block\, Level 7
DTSTART;TZID=Asia/Tokyo:20241204T130000
DTEND;TZID=Asia/Tokyo:20241204T131100
UID:siggraphasia_SIGGRAPH Asia 2024_sess117_papers_891@linklings.com
SUMMARY:FreeAvatar: Robust 3D Facial Animation Transfer by Learning an Exp
 ression Foundation Model
DESCRIPTION:Feng Qiu and Wei Zhang (Netease); Chen Liu (University of Quee
 nsland, Netease); Rudong An, Lincheng Li, Yu Ding, Changjie Fan, and Zhipe
 ng Hu (Netease); and Xin Yu (University of Queensland)\n\nVideo-driven 3D 
 facial animation transfer aims to drive avatars to reproduce the expressio
 ns of actors. Existing methods have achieved remarkable results by constra
 ining both geometric and perceptual consistency. However, geometric constr
 aints (like those designed on facial landmarks) are insufficient to captur
 e subtle emotions, while expression features trained on classification tas
 ks lack fine granularity for complex emotions. To address this, we propose
  \textbf{FreeAvatar}, a robust facial animation transfer method that relie
 s solely on our learned expression representation. Specifically, FreeAvata
 r consists of two main components: the expression foundation model and the
  facial animation transfer model. In the first component, we initially con
 struct a facial feature space through a face reconstruction task and then 
 optimize the expression feature space by exploring the similarities among 
 different expressions. Benefiting from training on the amounts of unlabele
 d facial images and re-collected expression comparison dataset, our model 
 adapts freely and effectively to any in-the-wild input facial images. In t
 he facial animation transfer component, we propose a novel Expression-driv
 en Multi-avatar Animator, which first maps expressive semantics to the fac
 ial control parameters of 3D avatars and then imposes perceptual constrain
 ts between the input and output images to maintain expression consistency.
  To make the entire process differentiable, we employ a trained neural ren
 derer to translate rig parameters into corresponding images. Furthermore, 
 unlike previous methods that require separate decoders for each avatar, we
  propose a dynamic identity injection module that allows for the joint tra
 ining of multiple avatars within a single network. The comparisons show th
 at our method achieves prominent performance even without introducing any 
 geometric constraints, highlighting the robustness of our FreeAvatar.\n\nR
 egistration Category: Full Access, Full Access Supporter\n\nLanguage Forma
 t: English Language\n\nSession Chair: Jungdam Won (Seoul National Universi
 ty)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_891&sess=sess117
END:VEVENT
END:VCALENDAR