Research map / 3D vision / Human pose estimation
From Pairs to Sequences: Track-Aware Policy Gradients for Keypoint Detection
Yepeng Liu, Hao Li, Liwen Yang, Fangzhen Li, Xudi Ge, Yuliang Gu, Kuang Gao, Bing Wang, Guang Chen, Hangjun Ye, Yongchao Xu
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 sequence-level rewards instead of pairwise (ablation) | TraqPoint · MegaDepth-1500 / KITTI Seq 01-03 | AUC@5 / AKTL: AUC@5 53.3 -> 55.8; AKTL 4.3 -> 6.6 | Improved |
| use track reward instead of basic match reward (ablation) | TraqPoint · MegaDepth-1500 / KITTI Seq 01-03 | AUC@5 / AKTL: 49.7 -> 53.3 AUC@5; 2.8 -> 4.3 AKTL | Improved |
| add rank reward (ablation) | TraqPoint · MegaDepth-1500 / KITTI Seq 01-03 | AUC@5 / AKTL: 52.6 -> 55.8 AUC@5; 4.0 -> 6.6 AKTL | Improved |
| add distinctiveness reward (ablation) | TraqPoint · MegaDepth-1500 / KITTI Seq 01-03 | AUC@5 / AKTL: 54.6 -> 55.8 AUC@5; 5.9 -> 6.6 AKTL | Improved |
| add RL training over supervised training (ablation) | TraqPoint · MegaDepth-1500 / KITTI Seq 01-03 | AUC@5 / AKTL: 52.0 -> 55.8 AUC@5; 3.8 -> 6.6 AKTL | Improved |
| use DINOv3-ConvNeXt instead of DINOv2-ViT (ablation) | TraqPoint · MegaDepth-1500 / KITTI Seq 01-03 | AUC@5 / AKTL: 54.8 -> 55.8 AUC@5; 6.4 -> 6.6 AKTL | Improved |
| use DINOv3-ConvNeXt instead of ResNet-50 (ablation) | TraqPoint · MegaDepth-1500 / KITTI Seq 01-03 | AUC@5 / AKTL: 54.5 -> 55.8 AUC@5; 6.1 -> 6.6 AKTL | Improved |
| lengthen training sequence beyond 5 frames (ablation) | TraqPoint · MegaDepth-1500 / KITTI Seq 01-03 | AUC@5 / AKTL | Hurt |
| use TraqPoint instead of RDD | TraqPoint · MegaDepth-1500 | AUC@5: 51.9 -> 55.8 | Improved |
| use TraqPoint instead of RDD | TraqPoint · ETH Madrid Metropolis | Reproj Err: 1.03px vs RootSIFT 0.60px | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- SP+SG — SPSG: Self-Supervised Photometric Scene Generation From RGB-D Scans (CVPR 2021) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Detect by Track: Making Detector-Free Matcher Trackable (ECCV 2026 · similarity 0.87)
- Towards Complementary Keypoint Detection via Mixture of Detectors (NeurIPS 2026 · similarity 0.86)
- Key-Grid: Unsupervised 3D Keypoints Detection using Grid Heatmap Features (NeurIPS 2024 · similarity 0.84)
- Learning to Produce Semi-dense Correspondences for Visual Localization (CVPR 2024 · similarity 0.83)
- DriveTrack: A Benchmark for Long-Range Point Tracking in Real-World Videos (CVPR 2024 · similarity 0.83)
- Back to 3D: Few-Shot 3D Keypoint Detection with Back-Projected 2D Features (CVPR 2024 · similarity 0.83)
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.
