Research map / 3D vision / Human pose estimation
Superman: Unifying Skeleton and Vision for Human Motion Perception and Generation
Xinshun Wang, Peiming Li, Ziyi Wang, Zhongbin Fang, Zhichao Deng, Songtao Wu, Jason Li, Mengyuan Liu
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 |
|---|---|---|---|
| use vision-guided fused tokenizer (0.5/0.5) instead of skeleton-only (ablation) | Superman · Human3.6M pose estimation | N-MPJPE (mm): 47.8 -> 44.9 | Improved |
| use vision-guided fused tokenizer instead of visual-only (ablation) | Superman · Human3.6M pose estimation | N-MPJPE (mm): 51.3 -> 44.9 | Improved |
| use balanced commitment weights 0.5/0.5 instead of 0.7/0.3 (ablation) | Superman tokenizer · Tokenizer reconstruction | reconstruction error (mm): 5.3 -> 4.7 | Improved |
| train one unified model instead of specialized per-task models (ablation) | Superman · Human3.6M pose estimation | MPJPE (mm): 46.5 -> 44.9 | Improved |
| train one unified model instead of specialized per-task models (ablation) | Superman · Human3.6M motion prediction | MPJPE (mm): 27.3 -> 26.1 | Improved |
| train one unified model instead of specialized per-task models (ablation) | Superman · Human3.6M motion in-betweening | MPJPE (mm): 33.1 -> 30.6 | Improved |
| scale MLLM from 3B to 7B (ablation) | Superman · Human3.6M PE and MP | MPJPE (mm) | Improved |
| increase codebook capacity (e.g. 4096x1024 to 4096x2048) (ablation) | Superman · Human3.6M PE and MP | MPJPE (mm) | Improved |
| add MAFT module (<0.2% extra params) for pose estimation (ablation) | Superman · Human3.6M pose estimation | N-MPJPE (mm): 44.90 -> 39.41 | Improved |
| Superman (no MAFT) vs Human-in-Context on pose estimation | Superman · Human3.6M pose estimation | N-MPJPE (mm): 44.77 -> 44.90 | Mixed |
| zero-shot transfer from Human3.6M to 3DPW | Superman · 3DPW motion prediction | MPJPE avg (mm): 140.71 -> 62.05 | Improved |
| zero-shot transfer from Human3.6M to 3DPW | Superman · 3DPW motion in-betweening | MPJPE avg (mm): 103.97 -> 60.68 | Improved |
Closest papers by meaning
- SAMoR: Motion Modelling for Articulated Objects of Any Skeleton and Topology (NeurIPS 2026 · similarity 0.89)
- MoVie: Broaden Your Views with Human Motion for Action Detection (CVPR 2026 · similarity 0.88)
- MotionBERT: A Unified Perspective on Learning Human Motion Representations (ICCV 2023 · similarity 0.88)
- Motion Keyframe Interpolation for Any Human Skeleton using Point Cloud-based Human Motion Data Homogenisation (ECCV 2024 · similarity 0.88)
- Exploring Vision Transformers for 3D Human Motion-Language Models with Motion Patches (CVPR 2024 · similarity 0.87)
- Universal Skeleton Understanding via Differentiable Rendering and MLLMs (ICML 2026 · 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.
