Research map / 3D vision / Human pose estimation
Learning Trajectory Dependencies for Human Motion Prediction
Wei Mao, Miaomiao Liu, Mathieu Salzmann, Hongdong Li
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 DCT trajectory encoding instead of raw joint angles (ablation) | DCT-GCN motion predictor · H3.6M 4-action avg @400ms | Euler angle error: 0.74 -> 0.71 | Improved |
| use DCT trajectory encoding instead of raw 3D coordinates (ablation) | DCT-GCN motion predictor · H3.6M 4-action avg @400ms | MPJPE (mm): 43.1 -> 38.4 | Improved |
| add input padding with last observed frame (ablation) | DCT-GCN motion predictor · H3.6M 4-action avg @400ms | Euler angle error: 0.86 -> 0.71 | Improved |
| add input padding with last observed frame (ablation) | DCT-GCN motion predictor · H3.6M 4-action avg @400ms | MPJPE (mm): 50.4 -> 38.4 | Improved |
| add residual connections (ablation) | DCT-GCN motion predictor · H3.6M 4-action avg @400ms | Euler angle error: 0.83 -> 0.71 | Improved |
| add residual connections (ablation) | DCT-GCN motion predictor · H3.6M 4-action avg @400ms | MPJPE (mm): 50.2 -> 38.4 | Improved |
| learn graph connectivity instead of pre-defined kinematic tree adjacency (ablation) | DCT-GCN motion predictor · H3.6M 4-action avg @400ms | Euler angle error: 0.88 -> 0.71 | Improved |
| learn graph connectivity instead of pre-defined kinematic tree adjacency (ablation) | DCT-GCN motion predictor · H3.6M 4-action avg @400ms | MPJPE (mm): 71.0 -> 38.4 | Improved |
| use GCN with learned connectivity instead of fully-connected network (ablation) | DCT-GCN motion predictor · H3.6M 4-action avg @400ms | Euler angle error: 0.76 -> 0.71 | Improved |
| use GCN with learned connectivity instead of fully-connected network (ablation) | DCT-GCN motion predictor · H3.6M 4-action avg @400ms | MPJPE (mm): 42.8 -> 38.4 | Improved |
| train directly on 3D coordinates instead of angles then converting to 3D | DCT-GCN motion predictor · H3.6M 15-action avg @400ms | MPJPE (mm): 65.0 -> 61.3 | Improved |
| compare against AGED w/adv at 1000ms in angle space | DCT-GCN motion predictor · H3.6M long-term angle error @1000ms | Euler angle error: 1.09 -> 1.27 | Hurt |
Closest papers by meaning
- Motion Prediction Using Trajectory Cues (ICCV 2021 · similarity 0.93)
- Progressively Generating Better Initial Guesses Towards Next Stages for High-Quality Human Motion Prediction (CVPR 2022 · similarity 0.92)
- Space-Time-Separable Graph Convolutional Network for Pose Forecasting (ICCV 2021 · similarity 0.92)
- Learning Dynamic Relationships for 3D Human Motion Prediction (CVPR 2020 · similarity 0.91)
- Generating Smooth Pose Sequences for Diverse Human Motion Prediction (ICCV 2021 · similarity 0.91)
- SimpliHuMoN: Simplifying Human Motion Prediction (NeurIPS 2026 · 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.
