Research map / 3D vision / Human pose estimation
Detect-and-Track: Efficient Pose Estimation in Videos
Read the paperCheck 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 |
|---|---|---|---|
| Add spatiotemporal 3D convolutions with center initialization | Mask R-CNN on PoseTrack validation · PoseTrack validation | mAP: 19.9 -> 22.6 | Helped |
| Add spatiotemporal 3D convolutions with center initialization | Mask R-CNN on PoseTrack validation · PoseTrack validation | MOTA: 14.1 -> 15.4 | Helped |
| Use center rather than mean initialization for the 3D kernels (ablation) | 3D Mask R-CNN · PoseTrack validation | mAP: 20.4 -> 22.6 | Helped |
| Use center rather than mean initialization for the 3D kernels (ablation) | 3D Mask R-CNN · PoseTrack validation | MOTA: 13.4 -> 15.4 | Helped |
| Raise the detection threshold from 0.0 to 0.95 (ablation) | Mask R-CNN tracker · PoseTrack validation | MOTA: 53.6 -> 55.2 | Helped |
| Raise the detection threshold from 0.0 to 0.95 (ablation) | Mask R-CNN tracker · PoseTrack validation | mAP: 64.1 -> 60.6 | Hurt |
| Replace Hungarian matching with greedy matching (ablation) | Detect-and-Track · PoseTrack validation | MOTA: 55.2 -> 55.1 | Hurt |
| Use pose PCK instead of bounding-box overlap as matching cost (ablation) | Detect-and-Track · PoseTrack validation | MOTA: 55.2 -> 54.2 | Hurt |
| Use CNN cosine distance instead of bounding-box overlap as matching cost (ablation) | Detect-and-Track · PoseTrack validation | MOTA: 55.2 -> 55.4 | No clear effect |
| Replace hand-crafted matching cost with a learned LSTM metric (ablation) | Detect-and-Track · PoseTrack validation | MOTA: 55.2 -> 48.4 | Hurt |
| Replace frame-level 2D pose estimation with 3D clip-based estimation (ablation) | Mask R-CNN with COCO initialization · PoseTrack validation | mAP: 19.9 -> 22.6 | Helped |
| Use perfect keypoints instead of predicted keypoints (ablation) | Detect-and-Track · PoseTrack validation | MOTA: 55.2 -> 78.4 | Helped |
Closest papers by meaning
- PoseTrack: A Benchmark for Human Pose Estimation and Tracking (CVPR 2018 · similarity 0.93)
- 15 Keypoints Is All You Need (CVPR 2020 · similarity 0.92)
- Optimizing Human Pose Estimation Through Focused Human and Joint Regions (AAAI 2025 · similarity 0.90)
- CoMotion: Concurrent Multi-person 3D Motion (ICLR 2025 · similarity 0.88)
- Motions as Queries: One-Stage Multi-Person Holistic Human Motion Capture (CVPR 2025 · similarity 0.88)
- ProbPose: A Probabilistic Approach to 2D Human Pose Estimation (CVPR 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.