Research map / 3D vision / Human pose estimation
Self-supervised Keypoint Correspondences for Multi-Person Pose Estimation and Tracking in Videos
Umer Rafi, Andreas Doering, Bastian Leibe, Juergen Gall
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 keypoint correspondences instead of optical flow for track association (ablation) | pose tracking · PoseTrack 2017 val (GT boxes) | MOTA: 68.4 -> 70.5 | Improved |
| use keypoint correspondences instead of person re-ID for association (ablation) | pose tracking · PoseTrack 2017 val (GT boxes) | MOTA: 67.1 -> 70.5 | Improved |
| use keypoint correspondences instead of OKS similarity (ablation) | pose tracking · PoseTrack 2017 val (detected boxes) | MOTA: 60.4 -> 67.9 | Improved |
| use keypoint correspondences instead of optical flow (ablation) | pose tracking · PoseTrack 2017 val (GT boxes) | IDSW: 4419 -> 3583 | Improved |
| add the refinement module for affinity maps (ablation) | correspondence tracking · PoseTrack 2017 val | MOTA: 66.9 -> 67.9 | Improved |
| add duplicate removal (ablation) | correspondence tracking · PoseTrack 2017 val | MOTA: 64.5 -> 67.9 | Improved |
| add recovery of missed detections via correspondences (ablation) | top-down pose estimator · PoseTrack 2018 val (threshold 0.3) | mAP: 77.7 -> 79.2 | Improved |
| add recovery of missed detections via correspondences (ablation) | top-down pose estimator · PoseTrack 2018 val (threshold 0.3) | MOTA: 67.6 -> 68.8 | Improved |
| raise joint detection threshold from 0.0 to 0.3 (ablation) | pose tracking · PoseTrack 2018 val (detected poses only) | mAP and MOTA: mAP 80.1 -> 77.7; MOTA 48.1 -> 67.6 | Mixed |
| vary tau_corr from 0.1 to 0.5 (ablation) | correspondence tracking · PoseTrack 2017 val | MOTA: 67.9 -> 67.8 | No clear effect |
| add track merging post-processing | correspondence tracking · PoseTrack 2017 val | MOTA: 67.9 -> 68.3 | Improved |
| add track merging post-processing | correspondence tracking · PoseTrack 2018 val | MOTA: 68.8 -> 69.1 | Improved |
Closest papers by meaning
- Detect-and-Track: Efficient Pose Estimation in Videos (CVPR 2018 · similarity 0.93)
- Temporal Keypoint Matching and Refinement Network for Pose Estimation and Tracking (ECCV 2020 · similarity 0.91)
- 15 Keypoints Is All You Need (CVPR 2020 · similarity 0.91)
- PoseTrack21: A Dataset for Person Search, Multi-Object Tracking and Multi-Person Pose Tracking (CVPR 2022 · similarity 0.90)
- Simple Pose: Rethinking and Improving a Bottom-up Approach for Multi-Person Pose Estimation (AAAI 2020 · similarity 0.90)
- CoMotion: Concurrent Multi-person 3D Motion (ICLR 2025 · 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.
