Search papers, labs, and topics across Lattice.
This paper introduces a Prior-Guided Residual Flow Matching framework that enhances 3D multi-person motion prediction by integrating a Deterministic Coarse Prior to maintain structural stability and a Dynamic Cross-Interaction mechanism for improved inter-agent communication. By reformulating the generative process to focus on motion residuals, the method effectively preserves kinematic coherence while addressing the challenges of noise-dominated integration. Extensive experiments validate the approach, achieving state-of-the-art accuracy across various datasets, underscoring its effectiveness in modeling complex multi-person interactions.
Achieving state-of-the-art accuracy in 3D multi-person motion prediction hinges on a novel framework that balances structural stability with dynamic inter-agent interactions.
3D multi-person motion prediction requires modeling both individual kinematics and inter-person interactions. While Flow Matching is effective for multi-hypothesis generation to improve prediction accuracy, directly predicting skeletal sequences from pure noise often compromises structural consistency and introduces unreliable cross-agent interactions during early noise-dominated integration steps. To address this, we propose a Prior-Guided Residual Flow Matching framework. First, a Deterministic Coarse Prior (DCP) establishes a kinematic anchor, formulating the generative process as a conditional flow over motion residuals to simplify the generative objective and preserve structural stability. Second, a Dynamic Cross-Interaction (DCI) mechanism temporally synchronizes inter-agent message-passing with the integration progress, ensuring the extraction of reliable social contexts and improving multi-person motion fidelity. Finally, a decoupled joint-motion architecture with bidirectional fusion effectively preserves fine-grained kinematic coherence. Extensive experiments demonstrate that our approach achieves state-of-the-art prediction accuracy across multiple datasets. Code is available at https://github.com/Wei-Wei-a/Residual-Flow-Matching-with-Dynamic-Cross-Interaction-for-3D-Multi-Person-Motion-Prediction.
