DSG-Net: Learning Disentangled Structure and Geometry for 3D Shape Generation
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
DescriptionWe present DSG-Net, a novel deep generative model that learns to represent and generate 3D shapes in disentangled latent spaces of geometry and structure while considering their synergy to ensure plausibility of generation. Our method also enables disentangled control of geometry and structure in shape generation, supporting novel applications such as interpolation of geometry (structure) while keeping structure (geometry) unchanged.