DeepPhase: Periodic Autoencoders for Learning Motion Phase Manifolds
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 the Periodic Autoencoder that can learn the spatial-temporal structure of body movements from unstructured motion data. The network produces a multi-dimensional phase manifold that helps enhance neural character controllers and motion matching for a variety of tasks, including diverse locomotion, style-based movements, dancing to music, or football dribbling.