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:20250110T023312Z
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:Technical Papers\n\nFeng Qiu and Wei Zhang (Netease); Chen Liu
  (University of Queensland, Netease); Rudong An, Lincheng Li, Yu Ding, Cha
 ngjie Fan, and Zhipeng Hu (Netease); and Xin Yu (University of Queensland)
 \n\nVideo-driven 3D facial animation transfer aims to drive avatars to rep
 roduce the expressions of actors. Existing methods have achieved remarkabl
 e results by constraining both geometric and perceptual consistency. Howev
 er, geometric constraints (like those designed on facial landmarks) are in
 sufficient to capture subtle emotions, while expression features trained o
 n classification tasks lack fine granularity for complex emotions. To addr
 ess this, we propose \textbf{FreeAvatar}, a robust facial animation transf
 er method that relies solely on our learned expression representation. Spe
 cifically, FreeAvatar consists of two main components: the expression foun
 dation model and the facial animation transfer model. In the first compone
 nt, we initially construct a facial feature space through a face reconstru
 ction task and then optimize the expression feature space by exploring the
  similarities among different expressions. Benefiting from training on the
  amounts of unlabeled facial images and re-collected expression comparison
  dataset, our model adapts freely and effectively to any in-the-wild input
  facial images. In the facial animation transfer component, we propose a n
 ovel Expression-driven Multi-avatar Animator, which first maps expressive 
 semantics to the facial control parameters of 3D avatars and then imposes 
 perceptual constraints between the input and output images to maintain exp
 ression consistency. To make the entire process differentiable, we employ 
 a trained neural renderer to translate rig parameters into corresponding i
 mages. Furthermore, unlike previous methods that require separate decoders
  for each avatar, we propose a dynamic identity injection module that allo
 ws for the joint training of multiple avatars within a single network. The
  comparisons show that our method achieves prominent performance even with
 out introducing any geometric constraints, highlighting the robustness of 
 our FreeAvatar.\n\nRegistration Category: Full Access, Full Access Support
 er\n\nLanguage Format: English Language\n\nSession Chair: Jungdam Won (Seo
 ul National University)
URL:https://asia.siggraph.org/2024/program/?id=papers_891&sess=sess117
END:VEVENT
END:VCALENDAR