Research map / Video / Video generation
FlowerDance: MeanFlow for Efficient and Refined 3D Dance Generation
Kaixing Yang, Xulong Tang, Ziqiao Peng, Xiangyue Zhang, Chubin Chen, xukun zhou, Puwei Wang, Hongyan Liu, Jun He
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 |
|---|---|---|---|
| replace rectified flow with MeanFlow (20 steps) (ablation) | FlowerDance · FineDance | FIDk: 49.59 -> 29.73 | Improved |
| reduce MeanFlow steps from 20 to 10 (ablation) | FlowerDance · FineDance | FIDk: 29.73 -> 26.17 | No clear effect |
| reduce MeanFlow steps from 20 to 5 (ablation) | FlowerDance · FineDance | FIDg: 19.59 -> 55.96 | Hurt |
| add Physical Consistency Constraint (position+velocity losses) (ablation) | MeanFlow · FineDance | FIDk: NaN -> 29.73 | Improved |
| add position and velocity losses to reconstruction loss (ablation) | PCC · FineDance | FIDk: 37.45 -> 29.73 | Improved |
| add velocity loss on top of reconstruction+position (ablation) | PCC · FineDance | FIDk: 25.86 -> 29.73 | Mixed |
| use BiMamba instead of Mamba (ablation) | FlowerDance · FineDance | FIDk: 39.40 -> 29.73 | Improved |
| use BiMamba instead of Transformer (NAR) (ablation) | FlowerDance · FineDance | all metrics: NaN -> 29.73 | Improved |
| use BiMamba instead of inpainting Transformer (II) (ablation) | FlowerDance · FineDance | FIDk: 23.29 -> 29.73 | Mixed |
| use element-wise addition instead of cross-attention fusion (ablation) | FlowerDance · FineDance | FIDk: 32.40 -> 29.73 | Improved |
| add time-decayed soft mask in motion editing (ablation) | FlowerDance · motion editing | boundary continuity (qualitative) | Improved |
| FlowerDance vs MEGA on inference speed | FlowerDance · FineDance efficiency | FPS@10^4: 340 -> 4145 | 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 5 of 7 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
- MEGADance: Mixture-of-Experts Architecture for Genre-Aware 3D Dance Generation (NeurIPS 2025 · 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)
- Lodge: A Coarse to Fine Diffusion Network for Long Dance Generation Guided by the Characteristic Dance Primitives (CVPR 2024 · similarity 0.87)
- Controllable 3D Dance Generation Using Diffusion-Based Transformer U-Net (AAAI 2025 · similarity 0.87)
- DanceCamera3D: 3D Camera Movement Synthesis with Music and Dance (CVPR 2024 · similarity 0.87)
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.
