Research map / Video / Video segmentation and tracking
Spatiotemporal-Untrammelled Mixture of Experts for Multi-Person Motion Prediction
Zheng Yin, Chengjian Li, Xiangbo Shu, Meiqi Cao, Rui Yan, Jinhui Tang
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 a single TT expert to the encoder/decoder baseline (ablation) | baseline · CMU-Mocap (UMPM) | avg JPE (mm): 111.1 -> 98.1 | Improved |
| add only the ST expert (ablation) | baseline · CMU-Mocap (UMPM) | avg JPE (mm): 111.1 -> 104.5 | Improved |
| add only the ST expert (0.2s horizon) (ablation) | baseline · CMU-Mocap (UMPM) | JPE at 0.2s (mm): 35.7 -> 37.1 | Hurt |
| combine four heterogeneous experts instead of one expert type (ablation) | ST-MoE · CMU-Mocap (UMPM) | avg JPE (mm): 98.1 -> 95.0 | Improved |
| combine four heterogeneous experts instead of one expert type (ablation) | ST-MoE · CMU-Mocap (UMPM) | avg APE (mm): 66.4 -> 65.4 | Improved |
| activate all four experts instead of fewer (top-k) (ablation) | ST-MoE · CMU-Mocap (UMPM) | avg JPE/APE | Improved |
| use a single MoE layer instead of stacking more (ablation) | ST-MoE · CMU-Mocap (UMPM) | avg JPE/APE | Improved |
| replace attention with bidirectional Mamba experts vs IAFormer | ST-MoE · CMU-Mocap (UMPM) | training speed / parameters: 3.6x faster, 41.38% fewer parameters | Improved |
| ST-MoE vs IAFormer on Mix1 (6 persons) JPE | ST-MoE · Mix1 | avg JPE (mm): 114 -> 110 | Improved |
| ST-MoE vs IAFormer on Mix2 (10 persons) JPE/APE | ST-MoE · Mix2 | avg JPE/APE (mm): 108 -> 107 JPE; 69 -> 69 APE | No clear effect |
Closest papers by meaning
- Efficient Multi-Person Motion Prediction by Lightweight Spatial and Temporal Interactions (ICCV 2025 · similarity 0.89)
- Diverse Human Motion Prediction Guided by Multi-level Spatial-Temporal Anchors (ECCV 2022 · similarity 0.87)
- Stochastic Multi-Person 3D Motion Forecasting (ICLR 2023 · similarity 0.87)
- Multi-level Motion Attention for Human Motion Prediction (IJCV 2021 · similarity 0.86)
- SimpliHuMoN: Simplifying Human Motion Prediction (NeurIPS 2026 · similarity 0.85)
- MotionMap: Representing Multimodality in Human Pose Forecasting (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.
