BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Asia/Hong_Kong
X-LIC-LOCATION:Asia/Hong_Kong
BEGIN:STANDARD
TZOFFSETFROM:+0800
TZOFFSETTO:+0800
TZNAME:HKT
DTSTART:19911015T033000
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20251218T030653Z
LOCATION:Meeting Room S221\, Level 2
DTSTART;TZID=Asia/Hong_Kong:20251216T164000
DTEND;TZID=Asia/Hong_Kong:20251216T165100
UID:siggraphasia_SIGGRAPH Asia 2025_sess131_papers_1934@linklings.com
SUMMARY:Motion In-Betweening for Densely Interacting Characters
DESCRIPTION:Xiaotang Zhang and Ziyi Chang (Durham University), Qianhui Men
  (University of Bristol), and Hubert Shum (Durham University)\n\nMotion in
 -betweening is the problem to synthesize movement between keyposes. Tradit
 ional research focused primarily on single characters. Extending them to d
 ensely interacting characters is highly challenging, as it demands precise
  spatial-temporal correspondence between the characters to maintain the in
 teraction, while creating natural transitions towards predefined keyposes.
  In this research, we present a method for long-horizon interaction in-bet
 weening that enables two characters to engage and respond to one another n
 aturally. \nTo effectively represent and synthesize interactions, we propo
 se a novel solution called Cross-Space In-Betweening, which models the int
 eractions of each character across different conditioning representation s
 paces. \nWe further observe that the significantly increased constraints i
 n interacting characters heavily limit the solution space, leading to degr
 aded motion quality and diminished interaction over time. \nTo enable long
 -horizon synthesis, we present two solutions to maintain long-term interac
 tion and motion quality, thereby keeping synthesis in the stable region of
  the solution space.\nWe first sustain interaction quality by identifying 
 periodic interaction patterns through adversarial learning. \nWe further m
 aintain the motion quality by learning to refine the drifted latent space 
 and prevent pose error accumulation. \nWe demonstrate that our approach pr
 oduces realistic, controllable, and long-horizon in-between motions of two
  characters with dynamic boxing and dancing actions across multiple keypos
 es, supported by extensive quantitative evaluations and user studies.\n\nR
 egistration Category: Full Access, Full Access Supporter\n\nSession Chair:
  Jungdam Won (Seoul National University)\n\n
END:VEVENT
END:VCALENDAR