.png){kind=icon}
.png){kind=icon}
Return
ArtGDance: music-driven 3D group dance generation framework
.png){kind=icon}
PRE
AI
Y
R
H
K
DOI：10.1007/s11042-026-21851-0.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Group dance is a complex and collaborative art form demanding synchronized motion and dynamic spatial coordination, which is intrinsically guided by music’s rhythm, tempo, and emotion. Compared to solo dance generation, group dance generation poses additional challenges: it requires not only generating musically aligned individual motions but also modeling complex inter-dancer interactions, maintaining spatial coherence across multiple performers, and avoiding collisions—all while preserving the artistic expressiveness of the choreography. In this paper, we address the task of music-driven group dance generation, aiming to synthesize natural, coherent, and musically aligned group movements. We propose ArtGDance, a novel framework comprising a Music Encoder, a Dance Decoder with dual temporal-spatial attention mechanisms, and a Dance Discriminator, jointly trained via a generative adversarial strategy augmented with a Kinetic Loss to enhance realism, synchronization, and diversity. To the best of our knowledge, this is the first work to apply Generative Adversarial Networks (GANs) specifically to the group dance generation task. Furthermore, we introduce a retrieval-based evaluation framework—the first of its kind for group dance generation—that aligns music and dance in a shared embedding space via contrastive learning, enabling comprehensive assessment beyond low-level motion statistics. Extensive experiments on the AIOZ-GDANCE dataset demonstrate that ArtGDance achieves competitive performance in both generation quality and artistic alignment. Project page: https://artgdance.github.io/
Keywords:
Computational creativity
Human motion generation
Music-to-Dance generation
Group dance generation
Generative adversarial training
Journal
IF:
3
Papers:
1.9W
Citations:
3.2W
Organization
Cited Papers