DiffusionNet: Discretization Agnostic Learning on Surfaces
Event Type
Technical Paper
Interest Areas
Research & Education
Presentation Types
Registration Categories
Full Conference Supporter
Full Conference
Virtual Conference Supporter
Virtual Conference
Exhibitor Additional Full Conference
Exhibitor Full Conference
DescriptionDiffusionNet is a general-purpose method for deep learning on meshes, points clouds, etc., via simple spatial diffusion. The networks generalize by construction across different samplings, resolutions, and representations of shapes and scale to high-resolution inputs. Spatial support is automatically optimized as a continuous parameter.