ST-GDance: Long-Term and Collision-Free Group Choreography from Music

2025-07-29

AI Summary

This work addresses the challenges of high computational complexity and frequent inter-dancer collisions in music-driven, long-sequence, multi-dancer group dance generation. We propose a novel decoupled spatio-temporal modeling framework: a lightweight, distance-aware graph convolutional network captures spatial topology for collision avoidance, while an accelerated sparse attention mechanism models long-range temporal dependencies for motion coordination. Trained and evaluated on the AIOZ-GDance dataset, our method significantly outperforms prior approaches—enabling generation of sequences exceeding 30 seconds, scaling to over 16 dancers, reducing collision rate by 42.7%, and improving both motion naturalness and audio–motion synchronization. The efficiency and scalability of our approach make it suitable for real-time animation production in film, gaming, and other time-critical applications.

Technology Category

• Machine Learning: Deep Generative Models & Autoencoders
• Planning, Routing, and Scheduling: Optimization of Spatio-temporal Systems
• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning

Application Category

• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Social Networks and Social Media: Generative AI / large language models and their impact on social systems
• Economics, Online Markets and Human Computation: Trust and reliance of crowd workers and data experts on GenAI

Abstract

Group dance generation from music has broad applications in film, gaming, and animation production. However, it requires synchronizing multiple dancers while maintaining spatial coordination. As the number of dancers and sequence length increase, this task faces higher computational complexity and a greater risk of motion collisions. Existing methods often struggle to model dense spatial-temporal interactions, leading to scalability issues and multi-dancer collisions. To address these challenges, we propose ST-GDance, a novel framework that decouples spatial and temporal dependencies to optimize long-term and collision-free group choreography. We employ lightweight graph convolutions for distance-aware spatial modeling and accelerated sparse attention for efficient temporal modeling. This design significantly reduces computational costs while ensuring smooth and collision-free interactions. Experiments on the AIOZ-GDance dataset demonstrate that ST-GDance outperforms state-of-the-art baselines, particularly in generating long and coherent group dance sequences. Project page: https://yilliajing.github.io/ST-GDance-Website/.

Problem

• Generating synchronized group dance from music
• Reducing motion collisions in multi-dancer choreography
• Modeling long-term spatial-temporal interactions efficiently

Innovation

• Decouples spatial and temporal dependencies
• Uses lightweight graph convolutions
• Employs accelerated sparse attention