Research map / 3D vision / 3D generation
DanceFormer: Music Conditioned 3D Dance Generation with Parametric Motion Transformer
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Represent motion as beat-synchronized key poses with predicted in-between curves rather than direct frame sequences (ablation) | DanceFormer · PhantomDance | NPSS: 14.55 vs 8.03 | Helped |
| Add Kinematic Propagation Module instead of linear layers with standard multi-head attention (ablation) | DanceFormer · PhantomDance | NPSS: 10.54 vs 8.03 | Helped |
| Add FK-IK kinematic propagation instead of a two-layer GNN (ablation) | DanceFormer · PhantomDance | NPSS: 10.34 vs 8.03 | Helped |
| Use learned local attention instead of global attention (ablation) | DanceFormer · PhantomDance | NPSS: 9.73 vs 8.03 | Helped |
| Use learned local attention instead of fixed local attention (ablation) | DanceFormer · PhantomDance | NPSS: 9.01 vs 8.03 | Helped |
| Use learned local attention instead of Gaussian-masked local attention (ablation) | DanceFormer · PhantomDance | NPSS: 8.93 vs 8.03 | Helped |
| Use the two-stage key-pose and motion-curve framework instead of direct frame prediction (ablation) | DanceFormer · PhantomDance | BC: 0.317 vs 0.785 | Helped |
| Add KPM instead of linear layers and standard multi-head attention (ablation) | DanceFormer · PhantomDance | BC: 0.637 vs 0.785 | Helped |
| Use FK-IK propagation instead of a two-layer GNN (ablation) | DanceFormer · PhantomDance | BC: 0.692 vs 0.785 | Helped |
| Use learned local attention instead of global attention (ablation) | DanceFormer · PhantomDance | BC: 0.737 vs 0.785 | Helped |
| Use learned local attention instead of fixed local attention (ablation) | DanceFormer · PhantomDance | BC: 0.763 vs 0.785 | Helped |
| Use learned local attention instead of Gaussian-masked local attention (ablation) | DanceFormer · PhantomDance | BC: 0.764 vs 0.785 | Helped |
Closest papers by meaning
- DanceCamera3D: 3D Camera Movement Synthesis with Music and Dance (CVPR 2024 · similarity 0.90)
- DanceAnyWay: Synthesizing Beat-Guided 3D Dances with Randomized Temporal Contrastive Learning (AAAI 2024 · similarity 0.90)
- Controllable 3D Dance Generation Using Diffusion-Based Transformer U-Net (AAAI 2025 · similarity 0.89)
- MEGADance: Mixture-of-Experts Architecture for Genre-Aware 3D Dance Generation (NeurIPS 2025 · similarity 0.88)
- POPDG: Popular 3D Dance Generation with PopDanceSet (CVPR 2024 · similarity 0.87)
- Bidirectional Autoregessive Diffusion Model for Dance Generation (CVPR 2024 · similarity 0.86)
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.