Research map / 3D vision / Human pose estimation
Combining Detection and Tracking for Human Pose Estimation in Videos
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 a 3D clip model to propagate detected person locations across neighboring frames (ablation) | HRNet pose estimation · PoseTrack2017 val | mAP: 77.7 -> 82.3 | Helped |
| Add the Video Tracking Pipeline and link overlapping pose tracklets (ablation) | 3D HRNet · PoseTrack2017 val | mAP: 82.3 -> 83.1 | Helped |
| Add spatial merge of pose hypotheses (ablation) | 3D HRNet with Video Tracking Pipeline and Baseline Merge · PoseTrack2017 val | mAP: 83.1 -> 83.5 | Helped |
| Add temporal merge of pose hypotheses (ablation) | 3D HRNet with Video Tracking Pipeline and Baseline Merge · PoseTrack2017 val | mAP: 83.1 -> 83.4 | Helped |
| Use both spatial and temporal merging (ablation) | 3D HRNet with Video Tracking Pipeline · PoseTrack2017 val | mAP: 83.1 -> 83.8 | Helped |
| Use both spatial and temporal merging (ablation) | 3D HRNet with Video Tracking Pipeline · PoseTrack2017 val | MOTA: 70.8 -> 71.6 | Helped |
| Place 3D temporal filters in the early stages rather than the last stage (ablation) | HRNet Clip Tracking Network · PoseTrack2017 val | mAP: 80.6 -> 81.1 | Helped |
| Place 3D temporal filters in the early stages rather than all stages (ablation) | HRNet Clip Tracking Network · PoseTrack2017 val | MOTA: 68.0 -> 70.0 | Helped |
| Increase clip length from 1 to 3 frames (ablation) | 3D HRNet Clip Tracking Network · PoseTrack2017 val | mAP | Helped |
| Increase clip length beyond 3 frames (ablation) | 3D HRNet Clip Tracking Network · PoseTrack2017 val | mAP | Helped |
| Increase stepsize and reduce the number of keyframes (ablation) | 3D HRNet video inference · PoseTrack2017 val | mAP | Hurt |
| Use the largest tested stepsize for faster inference (ablation) | 3D HRNet video inference · PoseTrack2017 val | mAP: mAP 78.9 and MOTA 67.2; baseline 2D HRNet mAP 77.7 and MOTA 65.6 | Mixed |
Closest papers by meaning
- PoseTrack: Joint Multi-Person Pose Estimation and Tracking (CVPR 2017 · similarity 0.93)
- ArtTrack: Articulated Multi-Person Tracking in the Wild (CVPR 2017 · similarity 0.92)
- Detect-and-Track: Efficient Pose Estimation in Videos (CVPR 2018 · similarity 0.92)
- Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 2025 · similarity 0.89)
- Motions as Queries: One-Stage Multi-Person Holistic Human Motion Capture (CVPR 2025 · similarity 0.88)
- CoMotion: Concurrent Multi-person 3D Motion (ICLR 2025 · 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.