Music-to-Dance Generation via Atomic Movements
Music-to-Dance Generation via Atomic Movements proposes a structure-aware framework that models dance choreography as a sequence of atomic movements. It constructs an atomic movement vocabulary through segmentation and clustering of large-scale dance data, and uses large language models for semantic annotation and refinement. The framework includes an atomic movement planning stage (predicting type, duration, and timing) and a completion stage (generating transition movements).
This method improves the structural coherence and controllability of generated dance by decomposing dance into interpretable atomic movements. Next steps could verify the generality of its atomic movement vocabulary and cross-dataset transferability.
This research demonstrates the potential of using large language models for motion semantic annotation, potentially shifting the music-to-motion generation field from continuous signal modeling to structured compositional modeling.
This technology can be applied to virtual idols, game character animation, dance teaching, and other scenarios, improving the controllability and quality of motion generation.
Future work could focus on the application of this method in real-time interactive dance generation or virtual human motion choreography, as well as the standardization and expansion of the atomic movement vocabulary.