Research map / 3D vision / Human pose estimation
Investigating Pose Representations and Motion Contexts Modeling for 3D Motion Prediction
Zhenguang Liu, Shuang Wu, Shuyuan Jin, Shouling Ji, Qi Liu, Shijian Lu, Li Cheng
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 Stiefel manifold instead of 3D coordinates representation (ablation) | AHMR · H3.6m | MAE at 400ms: 1.25 -> 0.91 | Improved |
| use Stiefel manifold instead of axis angles under L2 loss (ablation) | AHMR · H3.6m | MAE at 400ms: 1.04 -> 0.99 | Improved |
| use Stiefel manifold instead of quaternions under L2 loss (ablation) | AHMR · H3.6m | MAE at 400ms: 1.02 -> 0.99 | Improved |
| use geodesic loss instead of L2 loss with Stiefel manifold (ablation) | AHMR · H3.6m | MAE at 400ms: 0.99 -> 0.91 | Improved |
| use geodesic loss instead of L2 with axis angle representation (ablation) | AHMR · H3.6m | MAE at 400ms: 1.04 -> 1.09 | Hurt |
| use geodesic loss instead of L2 with quaternion representation (ablation) | AHMR · H3.6m | MAE at 400ms: 1.02 -> 1.10 | Hurt |
| train with forward kinematics loss instead of geodesic loss (ablation) | AHMR · H3.6m MPE | MPE at 400ms: 58.4 -> 23.1 | Improved |
| use forward kinematics loss with axis angle or quaternion representation (ablation) | AHMR · H3.6m MPE | MPE: Fails to Converge | Hurt |
| add left and right forget gates in the encoder (ablation) | AHMR · H3.6m | MAE at 400ms: 1.03 -> 0.91 | Improved |
| add the sequence level state g (ablation) | AHMR · H3.6m | MAE at 1000ms: 1.73 -> 1.58 | Improved |
| add spatial attention (ablation) | AHMR · H3.6m | MAE at 400ms: 0.95 -> 0.91 | Improved |
| add spatial and temporal attention (ablation) | AHMR · H3.6m | MAE at 400ms: 1.05 -> 0.91 | Improved |
Closest papers by meaning
- Context-Aware Human Motion Prediction (CVPR 2020 · similarity 0.91)
- Multi-level Motion Attention for Human Motion Prediction (IJCV 2021 · similarity 0.91)
- Deep Representation Learning for Human Motion Prediction and Classification (CVPR 2017 · similarity 0.91)
- Motion Prediction Using Trajectory Cues (ICCV 2021 · similarity 0.91)
- MotionMap: Representing Multimodality in Human Pose Forecasting (CVPR 2025 · similarity 0.88)
- SimpliHuMoN: Simplifying Human Motion Prediction (NeurIPS 2026 · similarity 0.88)
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.
