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:20241203T135800
DTEND;TZID=Asia/Tokyo:20241203T140900
UID:siggraphasia_SIGGRAPH Asia 2024_sess105_tog_107@linklings.com
SUMMARY:Identity-Preserving Face Swapping via Dual Surrogate Generative Mo
 dels
DESCRIPTION:Ziyao Huang and Fan Tang (Institute of Computing Technology, C
 hinese Academy of Sciences); Yong Zhang (Tencent); Juan Cao, Chengyu Li, S
 heng Tang, and Jintao Li (Institute of Computing Technology, Chinese Acade
 my of Sciences); and Tong-Yee Lee (National Cheng Kung University)\n\nIn t
 his study, we revisit the fundamental setting of face-swapping models and 
 reveal that only using implicit supervision for training leads to the diff
 iculty of advanced methods to preserve the source identity. We propose a n
 ovel reverse pseudo-input generation approach to offer supplemental data f
 or training face-swapping models, which addresses the aforementioned issue
 . Unlike the traditional pseudo-label-based training strategy, we assume t
 hat arbitrary real facial images could serve as the ground-truth outputs f
 or the face-swapping network and try to generate corresponding input <sour
 ce, target> pair data.  Specifically, we involve a source-creating surroga
 te that alters the attributes of the real image while keeping the identity
 , and a target-creating surrogate intends to synthesize attribute-preserve
 d target images with different identities. Our framework, which utilizes p
 roxy-paired data as explicit supervision to direct the face-swapping train
 ing process, partially fulfills a credible and effective optimization dire
 ction to boost the identity-preserving capability. We design explicit and 
 implicit adaption strategies to better approximate the explicit supervisio
 n for face swapping.\nQuantitative and qualitative experiments on FF++, FF
 HQ, and wild images show that our framework could improve the performance 
 of various face-swapping pipelines in terms of visual fidelity and ID pres
 erving. Furthermore, we display applications with our method on re-aging, 
 swappable attribute customization, cross-domain, and video face swapping. 
 Code is available under https://github.com/ICTMCG/CSCS.\n\nRegistration Ca
 tegory: Full Access, Full Access Supporter\n\nLanguage Format: English Lan
 guage\n\nSession Chair: Kfir Aberman (Decart AI)\n\n
URL:https://asia.siggraph.org/2024/program/?id=tog_107&sess=sess105
END:VEVENT
END:VCALENDAR