Research map / 3D vision / Human pose estimation
History Repeats Itself: Human Motion Prediction via Motion Attention
Wei Mao, Miaomiao Liu, Mathieu Salzmann
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 motion attention over sub-sequence keys/query instead of frame-wise attention (ablation) | LTD-style GCN predictor · H3.6M (3D) | MPJPE at 1000 ms: 136.7 -> 112.4 | Improved |
| use motion attention over sub-sequence keys/query instead of frame-wise attention (ablation) | LTD-style GCN predictor · H3.6M (3D) | MPJPE at 80 ms: 24.0 -> 10.8 | Improved |
| add motion attention to the predictor (Ours vs LTD-50-25, same 50 observed frames) | LTD-50-25 · H3.6M (3D, average) | MPJPE at 400 ms: 61.5 -> 58.3 | Improved |
| add motion attention to the predictor (Ours vs LTD-50-25) | LTD-50-25 · H3.6M (3D, average) | MPJPE at 1000 ms: 112.4 -> 112.1 | Improved |
| add motion attention on repetitive action | LTD-10-10 · H3.6M (3D, Walking) | MPJPE at 1000 ms: 70.7 -> 58.1 | Improved |
| add motion attention on non-repetitive action | LTD-10-10 · H3.6M (3D, Directions) | MPJPE at 1000 ms: 105.8 -> 106.5 | Hurt |
| add motion attention (angles, 8 sub-sequences) | LTD-10-10 · H3.6M (angles, average) | Euler angle error at 400 ms: 0.95 -> 0.94 | Improved |
| add motion attention (angles) | LTD-10-10 · H3.6M (angles, average) | Euler angle error at 1000 ms: 1.68 -> 1.65 | Improved |
| add motion attention on AMASS-BMLrub | LTD-10-25 · AMASS-BMLrub | MPJPE at 1000 ms: 75.2 -> 67.2 | Improved |
| add motion attention, zero-shot on 3DPW | LTD-10-25 · 3DPW | MPJPE at 1000 ms: 75.5 -> 73.7 | Improved |
| add motion attention on AMASS-BMLrub short-term | LTD-10-10 · AMASS-BMLrub | MPJPE at 80 ms: 10.3 -> 11.3 | Hurt |
| observe 100 frames instead of 50 at test time (ablation) | Ours · H3.6M (100 selected sequences with long-range repetition) | MPJPE at 1000 ms: 121.1 -> 116.9 | Improved |
Closest papers by meaning
- Multi-level Motion Attention for Human Motion Prediction (IJCV 2021 · similarity 0.97)
- Context-Aware Human Motion Prediction (CVPR 2020 · similarity 0.90)
- Investigating Pose Representations and Motion Contexts Modeling for 3D Motion Prediction (TPAMI 2022 · similarity 0.90)
- Progressively Generating Better Initial Guesses Towards Next Stages for High-Quality Human Motion Prediction (CVPR 2022 · similarity 0.89)
- Learning Trajectory Dependencies for Human Motion Prediction (ICCV 2019 · similarity 0.89)
- FutureHuman3D: Forecasting Complex Long-Term 3D Human Behavior from Video Observations (CVPR 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.
