Core Concepts
Innovative approach for high-quality multi-dancer choreography generation.
Stats
In group choreography, the dataset used comprises 16.7 hours of synchronized music and 3D multi-dancer motion data.
The Trajectory-Controllable Diffusion model was trained on 4 Nvidia 4090 GPUs for 1 day.
Batch sizes used for training the Dance-Beat Navigator varied based on the number of dancers.
Quotes
"Our method achieves state-of-the-art results in high-quality dance movement."
"Extensive experiments demonstrate the superiority of our approach over existing methods."