Research map / Video / Video generation
ClusterStyle: Modeling Intra-Style Diversity with Prototypical Clustering for Stylized Motion Generation
Kerui Chen, Jianrong Zhang, Ming Li, Zhonglong Zheng, Hehe Fan
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 |
|---|---|---|---|
| add multi-prototype clustering style encoder instead of learnable classification head with cross-entropy (ablation) | ClusterStyle · Stylized motion generation | SRA: 3.32% drop in SRA when replaced | Improved |
| add L_intra to L_inter (ablation) | ClusterStyle style encoder · HumanML3D+100STYLE stylized motion generation | SRA: 75.182 -> 78.101 | Improved |
| add L_inter to L_intra (ablation) | ClusterStyle style encoder · HumanML3D+100STYLE stylized motion generation | SRA: 67.746 -> 78.101 | Improved |
| use prototype contrastive L_intra instead of cross-entropy on learnable prototypes (ablation) | ClusterStyle style encoder · HumanML3D+100STYLE stylized motion generation | SRA: 65.465 -> 78.101 | Improved |
| add L_intra to L_inter (ablation) | ClusterStyle style encoder · HumanML3D+100STYLE stylized motion generation | FID: 1.112 -> 1.137 | Hurt |
| use cosine similarity instead of L2 (ablation) | ClusterStyle · HumanML3D+100STYLE stylized motion generation | SRA: 76.483 -> 78.101 | Improved |
| use cosine similarity instead of L1 (ablation) | ClusterStyle · HumanML3D+100STYLE stylized motion generation | FID: 1.322 -> 1.137 | Improved |
| increase number of global prototypes K_g (ablation) | ClusterStyle · HumanML3D+100STYLE stylized motion generation | R-Precision | Mixed |
| increase number of local prototypes K_l from 30 to 50 (ablation) | ClusterStyle · HumanML3D+100STYLE stylized motion generation | SRA | No clear effect |
| use ClusterStyle instead of StyleMotif | StyleMotif baseline · Motion style transfer | FID: 1.375 -> 0.768 | Improved |
| use ClusterStyle instead of BiFlow | BiFlow baseline · HumanML3D+100STYLE stylized motion generation | R-Precision (Top-3): 0.613 -> 0.708 | 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.
- StyleMotif — StyleMotif: Multi-Modal Motion Stylization using Style-Content Cross Fusion (ICCV 2025) · better on 7 of 8 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
- StyleMotif: Multi-Modal Motion Stylization using Style-Content Cross Fusion (ICCV 2025 · similarity 0.91)
- SMooDi: Stylized Motion Diffusion Model (ECCV 2024 · similarity 0.89)
- Generative Human Motion Stylization in Latent Space (ICLR 2024 · similarity 0.89)
- Arbitrary Motion Style Transfer with Multi-condition Motion Latent Diffusion Model (CVPR 2024 · similarity 0.88)
- Scalable Motion Style Transfer with Constrained Diffusion Generation (AAAI 2024 · similarity 0.86)
- StyleMaster: Stylize Your Video with Artistic Generation and Translation (CVPR 2025 · similarity 0.85)
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.
