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:20260817T171530Z
LOCATION:Hall B7 (1)\, B Block\, Level 7
DTSTART;TZID=Asia/Tokyo:20241206T095800
DTEND;TZID=Asia/Tokyo:20241206T100900
UID:siggraphasia_SIGGRAPH Asia 2024_sess141_papers_472@linklings.com
SUMMARY:Neural Garment Dynamic Super-Resolution
DESCRIPTION:Meng Zhang and Jun Li (Nanjing University of Science and Techn
 ology)\n\nAchieving efficient, high-fidelity, high-resolution garment simu
 lation is challenging due to its computational demands. Conversely, low-re
 solution garment simulation is more accessible and ideal for low-budget de
 vices like smartphones. In this paper, we introduce a lightweight, learnin
 g-based method for garment dynamic super-resolution, designed to efficient
 ly enhance high-resolution, high-frequency details in low-resolution garme
 nt simulations. Starting with low-resolution garment simulation and underl
 ying body motion, we utilize a mesh-graph-net to compute super-resolution 
 features based on coarse garment dynamics and garment-body interactions. T
 hese features are then used by a hyper-net to construct an implicit functi
 on of detailed wrinkle residuals for each coarse mesh triangle. Considerin
 g the influence of coarse garment shapes on detailed wrinkle performance, 
 we correct the coarse garment shape and predict detailed wrinkle residuals
  using these implicit functions. Finally, we generate detailed high-resolu
 tion garment geometry by applying the detailed wrinkle residuals to the co
 rrected coarse garment. Our method enables roll-out prediction by iterativ
 ely using its predictions as input for subsequent frames, producing fine-g
 rained wrinkle details to enhance the low-resolution simulation. Despite t
 raining on a small dataset, our network robustly generalizes to different 
 body shapes, motions, and garment types not present in the training data. 
 We demonstrate significant improvements over state-of-the-art alternatives
 , particularly in enhancing the quality of high-frequency, fine-grained wr
 inkle details.\n\nRegistration Category: Full Access, Full Access Supporte
 r\n\nLanguage Format: English Language\n\nSession Chair: Sheldon Andrews (
 École de Technologie Supérieure (ÉTS), McGill University)\n\n
URL:https://asia.siggraph.org/2024/program/?id=papers_472&sess=sess141
END:VEVENT
END:VCALENDAR