Research map / 3D vision / Human pose estimation
Efficient Online Multi-Person 2D Pose Tracking With Recurrent Spatio-Temporal Affinity Fields
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 |
|---|---|---|---|
| Replace each 7x7 filter with three 3x3 filters (ablation) | Model I · PoseTrack 2017 validation | mAP: 64.3 -> 66.3 | Helped |
| Replace each 7x7 filter with three 3x3 filters (ablation) | Model I · PoseTrack 2017 validation | fps: 10 -> 14 | Helped |
| Train on video sequences with synthetic motion augmentation (ablation) | Model II · PoseTrack 2017 validation | mAP: 56.6 -> 63.6 | Helped |
| Reduce first-stage depth from 128 to 64 channels (ablation) | Model II trained in Image Mode · PoseTrack 2017 validation | mAP: 56.6 -> 62.6 | Helped |
| Train the keypoint module with previous-frame heatmap feedback (ablation) | Model II · PoseTrack 2017 validation | mAP: 61.5 -> 63.6 | Helped |
| Use cross-linked limb TAF topology B rather than keypoint TAF topology A (ablation) | Model I · PoseTrack 2017 validation | MOTA: 58.5 -> 59.4 | Helped |
| Use cross-linked limb TAF topology B rather than keypoint TAF topology A (ablation) | Model II · PoseTrack 2017 validation | MOTA: 57.4 -> 58.4 | Helped |
| Remove current-frame spatial links and use topology C (ablation) | Model III · PoseTrack 2017 validation | mAP: 61.6 -> 55.2 | Hurt |
| Use cross-linked limb TAF topology instead of keypoint-to-same-keypoint TAFs (ablation) | STAF · PoseTrack 2017 validation | MOTA: Model I-A 58.5; Model I-B 59.4; Model II-A 57.4; Model II-B 58.4 | Helped |
| Use multi-scale rather than single-scale inference | Model II-B · PoseTrack 2017 validation | mAP: 64.6 -> 71.5 | Helped |
| Use multi-scale rather than single-scale inference | Model II-B · PoseTrack 2017 validation | fps: 27 -> 7 | Hurt |
| Replace TAF tracking with the Lucas-Kanade tracker (ablation) | Model I · PoseTrack videos at reduced frame rates | MOTA | Mixed |
Closest papers by meaning
- Realtime Multi-Person 2D Pose Estimation Using Part Affinity Fields (CVPR 2017 · similarity 0.91)
- ArtTrack: Articulated Multi-Person Tracking in the Wild (CVPR 2017 · similarity 0.89)
- CoMotion: Concurrent Multi-person 3D Motion (ICLR 2025 · similarity 0.87)
- SpatioTemporal Learning for Human Pose Estimation in Sparsely-Labeled Videos (AAAI 2025 · similarity 0.87)
- Video-Based Human Pose Regression via Decoupled Space-Time Aggregation (CVPR 2024 · similarity 0.86)
- Motions as Queries: One-Stage Multi-Person Holistic Human Motion Capture (CVPR 2025 · similarity 0.85)
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.