Research map / 3D vision / Human pose estimation
Normalized Human Pose Features for Human Action Video Alignment
Jingyuan Liu, Mingyi Shi, Qifeng Chen, Hongbo Fu, Chiew-Lan Tai
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 pose normalization (FK onto a condition skeleton) before embedding (ablation) | Ours (Pose Features) · Dense correspondence | hit ratio (%): 53.69 -> 74.95 | Improved |
| add pose normalization before embedding (ablation) | Ours (Pose Features) · Dense correspondence | MSE: 17.6033 -> 12.7106 | Improved |
| use adaptive triplet mining instead of fixed temporal windows (ablation) | Ours (Pose Features) · Dense correspondence | hit ratio (%): 41.92 -> 74.95 | Improved |
| use adaptive triplet mining instead of fixed temporal windows (ablation) | Ours (Pose Features) · Dense correspondence | MSE: 28.7598 -> 12.7106 | Improved |
| use fully connected embedding network instead of SemGCN (ablation) | Ours (Pose Features) · Dense correspondence | hit ratio (%): 61.05 -> 74.95 | Improved |
| use fully connected embedding network instead of PointNet (ablation) | Ours (Pose Features) · Dense correspondence | hit ratio (%): 68.96 -> 74.95 | Improved |
| train normalization and embedding networks separately rather than jointly (ablation) | Ours (Pose Features) · Dense correspondence | hit ratio (%): 25.76 -> 74.95 | Improved |
| do not add a temporal convolution window to the pose feature (ablation) | Ours (Pose Features) · Dense correspondence | hit ratio (%): 74.95 -> 44.81 | Hurt |
| use normalized poses with L2 distance instead of learned pose features | Ours · Dense correspondence | hit ratio (%): 74.95 -> 94.25 | Improved |
| replace direct-supervised 3D pose estimators with FK-based normalization for anthropometry robustness | Ours · Human3.6M (bone-length augmented) | Delta MPJPE (mm): 58.27 -> 53.35 | Mixed |
| use 3D pose-based Euler angles instead of 2D pose-based Euler angles | Euler angle feature · Dense correspondence | hit ratio (%): 33.20 -> 66.92 | Improved |
Closest papers by meaning
- Human Pose Estimation Using Global and Local Normalization (ICCV 2017 · similarity 0.87)
- View-Invariant Probabilistic Embedding for Human Pose (ECCV 2020 · similarity 0.87)
- Capturing the Motion of Every Joint: 3D Human Pose and Shape Estimation with Independent Tokens (ICLR 2023 · similarity 0.87)
- Learning Human Pose Models from Synthesized Data for Robust RGB-D Action Recognition (IJCV 2019 · similarity 0.87)
- Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 2025 · similarity 0.87)
- Stitch, Contrast, and Segment: Learning a Human Action Segmentation Model Using Trimmed Skeleton Videos (AAAI 2025 · similarity 0.86)
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.
