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 S421\, Level 4
DTSTART;TZID=Asia/Hong_Kong:20251215T135300
DTEND;TZID=Asia/Hong_Kong:20251215T140400
UID:siggraphasia_SIGGRAPH Asia 2025_sess104_papers_1315@linklings.com
SUMMARY:PoissonNet: A Local-Global Approach for Learning on Surfaces
DESCRIPTION:Arman Maesumi and Tanish Makadia (Brown University), Thibault 
 Groueix and Vladimir Kim (Adobe Research), Daniel Ritchie (Brown Universit
 y), and Noam Aigerman (University of Montreal)\n\nMany network architectur
 es exist for learning on meshes, yet their constructions entail delicate t
 rade‑offs between difficulty learning high-frequency features, insufficien
 t receptive field, sensitivity to discretization, and inefficient computat
 ional overhead. Drawing from classic local-global approaches in mesh proce
 ssing, we introduce PoissonNet, a novel neural architecture that overcomes
  all of these deficiencies by formulating a local-global learning scheme, 
 which uses Poisson's equation as the primary mechanism for feature propaga
 tion. Our core network block is simple; we apply learned local feature tra
 nsformations in the gradient domain of the mesh, then solve a Poisson syst
 em to propagate scalar feature updates across the surface globally. Our lo
 cal‑global learning framework preserves the features's full frequency spec
 trum and provides a truly global receptive field, while remaining agnostic
  to mesh triangulation. Our construction is efficient, requiring far less 
 compute overhead than previous methods, which enables scalability---both i
 n the size of our datasets, and the size of individual training samples. T
 hese qualities are validated on various experiments where, compared to pre
 vious intrinsic architectures, we attain state-of-the-art performance on s
 emantic segmentation and parameterizing highly-detailed animated surfaces.
  Finally, as a central application of PoissonNet, we show its ability to l
 earn deformations, significantly outperforming all other architectures tha
 t learn on surfaces. Code will be made available upon publication.\n\nRegi
 stration Category: Full Access, Full Access Supporter\n\nSession Chair: Xu
 ejin Chen (University of Science and Technology of China)\n\n
END:VEVENT
END:VCALENDAR