BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Australia/Melbourne
X-LIC-LOCATION:Australia/Melbourne
BEGIN:DAYLIGHT
TZOFFSETFROM:+1000
TZOFFSETTO:+1100
TZNAME:AEDT
DTSTART:19721003T020000
RRULE:FREQ=YEARLY;BYMONTH=4;BYDAY=1SU
END:DAYLIGHT
BEGIN:STANDARD
DTSTART:19721003T020000
TZOFFSETFROM:+1100
TZOFFSETTO:+1000
TZNAME:AEST
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20240214T070250Z
LOCATION:Meeting Room C4.11\, Level 4 (Convention Centre)
DTSTART;TZID=Australia/Melbourne:20231215T160000
DTEND;TZID=Australia/Melbourne:20231215T161000
UID:siggraphasia_SIGGRAPH Asia 2023_sess139_papers_240@linklings.com
SUMMARY:A Locality-based Neural Solver for Optical Motion Capture
DESCRIPTION:Technical Papers\n\nXiaoyu Pan and Bowen Zheng (State Key Labo
 ratory of CAD & CG, Zhejiang University; ZJU-Tencent Game and Intelligent 
 Graphics Innovation Technology Joint Lab); Xinwei Jiang, Guanglong Xu, Xia
 nli Gu, and Jingxiang Li (Tencent Games Digital Content Technology Center)
 ; Qilong Kou (Tencent Technology (Shenzhen) Co., LTD); He Wang (University
  College London (UCL)); Tianjia Shao and Kun Zhou (State Key Laboratory of
  CAD & CG, Zhejiang University); and Xiaogang Jin (State Key Laboratory of
  CAD & CG, Zhejiang University; ZJU-Tencent Game and Intelligent Graphics 
 Innovation Technology Joint Lab)\n\nWe present a novel locality-based lear
 ning method for cleaning and solving optical motion capture data. Given no
 isy marker data, we propose a new heterogeneous graph neural network which
  treats markers and joints as different types of nodes, and uses graph con
 volution operations to extract the local features of markers and joints an
 d transform them to clean motions. To deal with anomaly markers (e.g. miss
 ing or with big tracking errors), the key insight is that a marker motion 
 show strong correlations with the motions of its immediate neighboring mar
 kers but less so with other markers, a.k.a. locality, which enables us to 
 fill missing markers (e.g. due to occlusion). Additionally, we also identi
 fy marker outliers due to tracking errors by investigating their accelerat
 ion profiles. Finally, we propose a training regime based on representatio
 n learning and data augmentation, by training the model on data with maski
 ng. The masking schemes aim to mimic the missing and noisy markers often o
 bserved in the real data. Finally, we show that our method achieves high a
 ccuracy on multiple metrics across various datasets. Extensive comparison 
 shows our method outperforms state-of-the-art methods in terms of predicti
 on accuracy of occluded marker position error by approximately 20%, which 
 leads to a further error reduction on the reconstructed joint rotations an
 d positions by 30%. The code and data for this paper are available at gith
 ub.com/localmocap/LocalMoCap.\n\nRegistration Category: Full Access\n\nSes
 sion Chair: Yuting Ye (Reality Labs Research, Meta)
URL:https://asia.siggraph.org/2023/full-program?id=papers_240&sess=sess139
END:VEVENT
END:VCALENDAR