Research map / 3D vision / 3D generation
HUMANISE: Language-conditioned Human Motion Generation in 3D Scenes
Zan Wang, Yixin Chen, Tengyu Liu, Yixin Zhu, Wei Liang, Siyuan Huang
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 target-object grounding loss (ablation) | HUMANISE model · HUMANISE, all actions | goal dist.: 1.008 -> 1.383 | Improved |
| add action classification loss (ablation) | HUMANISE model · HUMANISE, all actions | action score: 2.29$\pm$1.43 -> 3.59$\pm$1.38 | Improved |
| add both object-grounding and action-classification auxiliary losses (ablation) | HUMANISE model · HUMANISE, all actions | quality score: 1.97$\pm$0.98 -> 2.57$\pm$1.20 | Improved |
| fuse scene and language features using self-attention rather than direct concatenation (ablation) | HUMANISE model · HUMANISE, walk | goal dist.: 1.500 -> 1.370 | Improved |
| use Point Transformer instead of PointNet++ for scene encoding (ablation) | HUMANISE model · HUMANISE, walk | goal dist.: 1.444 -> 1.370 | Improved |
| use the full grounding-loss weight rather than reducing alpha_o (ablation) | HUMANISE model · HUMANISE, lie down | goal dist.: 0.444 -> 0.196; 0.306 -> 0.196; 0.211 -> 0.196 | Mixed |
| train using the full lie-down dataset rather than 50% of it (ablation) | HUMANISE model · HUMANISE, lie down | goal dist.: 0.177 -> 0.196 | Hurt |
| train using the full lie-down dataset rather than 10% of it (ablation) | HUMANISE model · HUMANISE, lie down | MPJPE: 248.82 -> 136.87 | Improved |
| pre-train the motion synthesis model on HUMANISE before fine-tuning on PROX | human motion synthesis model · PROX test split | translation: 7.65 -> 6.54 | Improved |
| provide ground-truth target position instead of relying on predicted language grounding | HUMANISE model · HUMANISE, all actions | goal dist.: 1.008 -> 0.207 | Improved |
| provide ground-truth action instead of parsing action from the language description | HUMANISE model · HUMANISE, all actions | goal dist.: 1.008 -> 1.406 | Mixed |
Later papers that compared with this method
3 later accepted papers report Humanise as a baseline in their results tables; 3 of them report a better number on most of the results they share with it.
- Learning Generatable Mutual Distance for Scene-Aware Human Motion Generation (ECCV 2026 · better on 5 of 5 shared results)
- SceMoS: Scene-Aware 3D Human Motion Synthesis by Planning with Geometry-Grounded Tokens (CVPR 2026 · better on 4 of 4 shared results)
- Generating Human Motion in 3D Scenes from Text Descriptions (CVPR 2024 · better on 2 of 2 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
- Generating Human Motion in 3D Scenes from Text Descriptions (CVPR 2024 · similarity 0.91)
- Move as You Say Interact as You Can: Language-guided Human Motion Generation with Scene Affordance (CVPR 2024 · similarity 0.90)
- InterDreamer: Zero-Shot Text to 3D Dynamic Human-Object Interaction (NeurIPS 2024 · similarity 0.90)
- GenZI: Zero-Shot 3D Human-Scene Interaction Generation (CVPR 2024 · similarity 0.90)
- The Language of Motion: Unifying Verbal and Non-verbal Language of 3D Human Motion (CVPR 2025 · similarity 0.90)
- HSI-GPT: A General-Purpose Large Scene-Motion-Language Model for Human Scene Interaction (CVPR 2025 · 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.
