Research map / Video / Video generation
Music-to-Dance Generation via Atomic Movements
Xinhao Cai, Yixuan Sun, Minghang Zheng, Qingchao Chen, Xin Jin, Song-Chun Zhu, Yang Liu
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| LLM-assisted in-group re-clustering instead of naive clustering (ablation) | atomic movement discovery · AIST++ | FID_k: 32.68 -> 25.26 | Improved |
| LLM-assisted re-clustering instead of re-clustering without LLM (ablation) | atomic movement discovery · AIST++ | R-precision: 23.3 -> 26.6 | Improved |
| use 125 base clusters instead of 100 (ablation) | atomic movement discovery · AIST++ | FID_k: 32.68 -> 34.57 | Hurt |
| use 75 base clusters instead of 100 (ablation) | atomic movement discovery · AIST++ | FID_k: 32.68 -> 33.24 | Hurt |
| add post-processing to atomic planning (ablation) | two-stage planner · AIST++ | FID_k: 25.26 -> 24.02 | Improved |
| use ground-truth atomic labels instead of planned labels (ablation) | two-stage framework · AIST++ | FID_k: 24.02 -> 21.59 | Improved |
| flexible re-creation of atomic movements instead of fixed concatenation (ablation) | dance completion · AIST++ | MultiModality: 1.43 -> 2.25 | Improved |
| flexible re-creation of atomic movements instead of fixed (ablation) | dance completion · AIST++ | FID_k: 24.13 -> 25.26 | Hurt |
| duration-nearest retrieval instead of random choice (ablation) | dance completion · AIST++ | FID_k: 27.94 -> 25.26 | Improved |
| duration-nearest retrieval instead of single candidate (ablation) | dance completion · AIST++ | Div_k: 6.42 -> 8.01 | Improved |
| two-stage atomic planning framework instead of Lodge | music-to-dance · AIST++ | R-precision: 18.2 -> 26.6 | Improved |
| two-stage framework instead of Bailando | music-to-dance · AIST++ | FID_k: 28.16 -> 25.26 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Lodge — Lodge++: High-Quality and Long Dance Generation With Robust Choreography Patterns (TPAMI 2025) · better on 4 of 4 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Text Dictates, Music Decorates: Energy-based Attention for Editable Dance Generation (ECCV 2026 · similarity 0.92)
- PhaseDance: Capturing Rhythm and Expressivity in Dance Modeling (NeurIPS 2026 · similarity 0.91)
- Dance Revolution: Long-Term Dance Generation with Music via Curriculum Learning (ICLR 2021 · similarity 0.90)
- DanceAnyWay: Synthesizing Beat-Guided 3D Dances with Randomized Temporal Contrastive Learning (AAAI 2024 · similarity 0.89)
- Bidirectional Autoregessive Diffusion Model for Dance Generation (CVPR 2024 · similarity 0.88)
- MEGADance: Mixture-of-Experts Architecture for Genre-Aware 3D Dance Generation (NeurIPS 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
