Scalable Group Choreography via Variational Phase Manifold Learning

Generating group dance motion from the music is a challenging task withseveral industrial applications. Although several methods have been proposed totackle this problem, most of them prioritize optimizing the fidelity in dancingmovement, constrained by predetermined dancer counts in datasets. Thislimitation impedes adaptability to real-world applications. Our study addressesthe scalability problem in group choreography while preserving naturalness andsynchronization. In particular, we propose a phase-based variational generativemodel for group dance generation on learning a generative manifold. Our methodachieves high-fidelity group dance motion and enables the generation with anunlimited number of dancers while consuming only a minimal and constant amountof memory. The intensive experiments on two public datasets show that ourproposed method outperforms recent state-of-the-art approaches by a largemargin and is scalable to a great number of dancers beyond the training data.