Research map / 3D vision / 3D generation
Open the Motion Door: Atomic Motion Decomposition and Recomposition for Open-Vocabulary Motion Generation
Ke Fan, Jiangning Zhang, Ran Yi, Jingyu Gong, Yabiao Wang, Yating Wang, Xin Tan, Chengjie Wang, Lizhuang Ma
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 TMA (motion-aligned contrastive text encoder) to the baseline (ablation) | Baseline atomic text-to-motion model · IDEA400 (out-of-domain) | R-Precision: 0.160 -> 0.380 | Improved |
| add CFF on top of TMA (ablation) | Baseline+TMA · IDEA400 (out-of-domain) | R-Precision: 0.380 -> 0.449 | Improved |
| add CFF on top of TMA (ablation) | Baseline+TMA · IDEA400 (out-of-domain) | FID: 0.844 -> 0.847 | Hurt |
| add CFF without TMA (feature-level fusion of atomic texts) (ablation) | Baseline · IDEA400 (out-of-domain) | R-Precision: 0.160 -> 0.170 | Improved |
| concatenate atomic texts with raw text (CFF*) instead of feature-level fusion (ablation) | Baseline · IDEA400 (out-of-domain) | FID: 0.898 -> 0.890 | No clear effect |
| add CFF (feature-level fusion) vs concatenation (ablation) | Baseline · IDEA400 (out-of-domain) | FID: 0.898 -> 0.886 | Improved |
| add TMA and CFF to the baseline (ablation) | Baseline · HumanML3D (in-domain) | FID: 0.140 -> 0.132 | Improved |
| add CFF to the baseline (ablation) | Baseline · HumanML3D (in-domain) | FID: 0.140 -> 0.141 | No clear effect |
| use textual decomposition into atomic texts instead of raw text (ablation) | text feature distribution (CLIP-B, t-SNE) · HumanML3D, IDEA400, Mixamo | domain gap (t-SNE overlap) | Improved |
| use full method vs MDM baseline on out-of-domain FID | Ours vs MDM · IDEA400 | FID: 0.821 (MDM) vs 0.847 (Ours) | Hurt |
Closest papers by meaning
- OpenT2M: No-frill Motion Generation with Open-source, Large-scale, High-quality Data (CVPR 2026 · similarity 0.92)
- VersatileMotion: A Unified Framework for Motion Synthesis and Comprehension (ECCV 2026 · similarity 0.91)
- MMM: Generative Masked Motion Model (CVPR 2024 · similarity 0.91)
- ReMoGPT: Part-Level Retrieval-Augmented Motion-Language Models (AAAI 2025 · similarity 0.90)
- SnapMoGen: Human Motion Generation from Expressive Texts (NeurIPS 2025 · similarity 0.90)
- OMG: Towards Open-vocabulary Motion Generation via Mixture of Controllers (CVPR 2024 · similarity 0.89)
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.
