Browse Teach Curate

Spatio-Temporal Multi-Subgraph Graph Convolutional Networks for Human Motion Prediction

How graph convolutional networks model the joint-to-joint and time dependencies of 3D skeletal motion by decoupling spatial and temporal modeling and transferring knowledge between them. Students learn why separating the two domains and enforcing consistency between them captures their complementarity, and how multiple subgraphs with a homogeneity constraint extract richer motion structure than a single graph.

Demerzel Text

2501.00317

STMS-GCN is a graph-convolutional network for 3D skeleton-based human motion prediction that decouples the modeling of temporal and spatial dependencies rather than treating them individually or blen…