Research map / 3D vision / 3D generation
MMG-VL: A Vision-Language Driven Approach for Multi-Person Motion Generation
Songyuan Yang, Wanrong Huang, Yinuo Liu, Zhang Ke-Di, Xihuai He, Shaowu Yang, Huibin Tan
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 Spatial-Semantic Grounding Loss (LSSG) to the planner (ablation) | MMG-VL · HumanVL (4-person ablation) | Environmental Alignment: 6.95 -> 7.80 | Improved |
| add LSSG to the planner (ablation) | MMG-VL · HumanVL (4-person ablation) | Multi-person Coordination: 8.75 -> 8.89 | No clear effect |
| add Lenv (SDF penetration loss) to the synthesizer (ablation) | MMG-VL · HumanVL (4-person ablation) | Penetration mean: 0.873 -> 0.251 | Improved |
| add Lenv to the synthesizer (ablation) | MMG-VL · HumanVL (4-person ablation) | Environmental Alignment: 5.82 -> 7.80 | Improved |
| add Lenv to the synthesizer (ablation) | MMG-VL · HumanVL (4-person ablation) | Diversity: 1.321 -> 1.310 | Hurt |
| add Lcoord (coordination loss) to the synthesizer (ablation) | MMG-VL · HumanVL (4-person ablation) | Multi-person Coordination: 6.23 -> 8.89 | Improved |
| add Lcoord to the synthesizer (ablation) | MMG-VL · HumanVL (4-person ablation) | Penetration mean: 0.288 -> 0.251 | Improved |
| add both Lenv and Lcoord (ablation) | MMG-VL · HumanVL (4-person ablation) | Environmental Alignment: 5.11 -> 7.80 | Improved |
| add both Lenv and Lcoord (ablation) | MMG-VL · HumanVL (4-person ablation) | Multi-person Coordination: 5.95 -> 8.89 | Improved |
| add both Lenv and Lcoord (ablation) | MMG-VL · HumanVL (4-person ablation) | Diversity: 1.328 -> 1.310 | Hurt |
| use MMG-VL CMS instead of MotionDiffuse | text-to-motion synthesis · HumanML3D | FID: 0.630 -> 0.501 | Improved |
| use MMG-VL instead of best HSI baseline with ground-truth 3D scenes | single-person scene-aware generation · HumanVL (single-person, vs HSI) | Penetration mean: 0.387 -> 0.283 | Improved |
Closest papers by meaning
- Unified Multi-Modal Interactive and Reactive 3D Motion Generation via Rectified Flow (ICLR 2026 · similarity 0.89)
- FantasyHSI: Video-Generation-Centric 4D Human Synthesis in Any Scene Through a Graph-Based Multi-Agent Framework (AAAI 2026 · similarity 0.88)
- HUMANISE: Language-conditioned Human Motion Generation in 3D Scenes (NeurIPS 2022 · similarity 0.88)
- Seamless Human Motion Composition with Blended Positional Encodings (CVPR 2024 · similarity 0.88)
- Sitcom-Crafter: A Plot-Driven Human Motion Generation System in 3D Scenes (ICLR 2025 · similarity 0.88)
- InterControl: Zero-shot Human Interaction Generation by Controlling Every Joint (NeurIPS 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.
