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:20251216T171300
DTEND;TZID=Asia/Hong_Kong:20251216T172400
UID:siggraphasia_SIGGRAPH Asia 2025_sess131_papers_1029@linklings.com
SUMMARY:StableMotion: Training Motion Cleanup Models with Unpaired Corrupt
 ed Data
DESCRIPTION:Yuxuan Mu (Simon Fraser University); Hung Yu Ling (Electronic 
 Arts); Yi Shi (Simon Fraser University); Ismael Baira Ojeda (Electronic Ar
 ts); Pengcheng Xi and Chang Shu (National Research Council Canada); Fabio 
 Zinno (Electronic Arts); and Xue Bin Peng (Simon Fraser University, NVIDIA
 )\n\nMotion capture (mocap) data often exhibits visually jarring artifacts
  due to inaccurate sensors and post-processing. Cleaning this corrupted da
 ta can require substantial manual effort from human experts, which can be 
 a costly and time-consuming process. Previous data-driven motion cleanup m
 ethods offer the promise of automating this cleanup process, but often req
 uire in-domain paired corrupted-to-clean training data. Constructing such 
 paired datasets requires access to high-quality, relatively artifact-free 
 motion clips, which often necessitates laborious manual cleanup. In this w
 ork, we present StableMotion, a simple yet effective method for training m
 otion cleanup models directly from unpaired corrupted datasets that need c
 leanup. The core component of our method is the introduction of motion qua
 lity indicators, which can be easily annotated— through manual labeling or
  heuristic algorithms—and enable training of quality-aware motion generati
 on models on raw motion data with mixed quality. At test time, the model c
 an be prompted to generate high-quality motions using the quality indicato
 rs. Our method can be implemented through a simple diffusion-based framewo
 rk, leading to a unified motion generate-discriminate model, which can be 
 used to both identify and fix corrupted frames. We demonstrate that our pr
 oposed method is effective for training motion cleanup models on raw mocap
  data in production scenarios by applying StableMotion to SoccerMocap, a 2
 45-hour soccer mocap dataset containing real-world motion artifacts. The t
 rained model effectively corrects a wide range of motion artifacts, reduci
 ng motion pops and frozen frames by 68% and 81%, respectively. On our benc
 hmark dataset, we further show that cleanup models trained with our method
  on unpaired corrupted data outperform state-of-the-art methods trained on
  clean or paired data, while also achieving comparable performance in pres
 erving the content of the original motion clips.\n\nRegistration Category:
  Full Access, Full Access Supporter\n\nSession Chair: Jungdam Won (Seoul N
 ational University)\n\n
END:VEVENT
END:VCALENDAR