InterCMDM: Block-Causal Diffusion for Autoregressive Human Interaction Generation

Researchers have introduced InterCMDM, a new framework designed to generate realistic human interactions by focusing on both individual movement patterns and coordinated joint actions. By utilizing block-causal diffusion, the model processes sequential motion data autoregressively to maintain temporal consistency between multiple people. This approach improves upon previous methods that struggled to balance long-range causality with the precise synchronization required for complex, multi-person activities.