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:20251218T030656Z
LOCATION:Meeting Room S221\, Level 2
DTSTART;TZID=Asia/Hong_Kong:20251218T105000
DTEND;TZID=Asia/Hong_Kong:20251218T110100
UID:siggraphasia_SIGGRAPH Asia 2025_sess153_papers_1923@linklings.com
SUMMARY:Learning Human Motion with Temporally Conditional Mamba
DESCRIPTION:Quang Nguyen and Tri Le (FPT AI Center); Baoru Huang (Universi
 ty of Liverpool, Imperial College London); Minh Nhat Vu (Vienna University
  of Technology); Ngan Le (University of Arkansas at Little Rock); Thieu Vo
  (National University of Singapore); and Anh Nguyen (Department of Compute
 r Science, University of Liverpool)\n\nLearning human motion based on a ti
 me-dependent input signal presents a challenging yet impactful task with v
 arious applications. The goal of this task is to generate or estimate huma
 n movement that consistently reflects the temporal patterns of conditionin
 g inputs. Existing methods typically rely on cross-attention mechanisms to
  fuse the condition with motion. However, this approach primarily captures
  global interactions and struggles to maintain step-by-step temporal align
 ment. To address this limitation, we introduce Temporally Conditional Mamb
 a, a new mamba-based model for human motion understanding. Our approach in
 tegrates conditional information into the recurrent dynamics of the Mamba 
 block, enabling better temporally aligned motion. To validate the effectiv
 eness of our method, we evaluate it on a variety of human motion tasks. Ex
 tensive experiments demonstrate that our model significantly improves temp
 oral alignment, motion realism, and condition consistency over state-of-th
 e-art approaches.\n\nRegistration Category: Full Access, Full Access Suppo
 rter\n\nSession Chair: Kai Wang (Simon Fraser University)\n\n
END:VEVENT
END:VCALENDAR