Research map / Detection and segmentation / Open-set and few-shot detection
UKPGAN: A General Self-Supervised Keypoint Detector
Yang You, Wenhai Liu, Yanjie Ze, Yong-Lu Li, Weiming Wang, Cewu Lu
Check 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 |
|---|---|---|---|
| add GAN-based sparsity control (ablation) | UKPGAN · KeypointNet airplane, chair, and table | IoU (%): 36.3 -> 68.8; 27.2 -> 36.2; 23.1 -> 34.7 | Improved |
| add GAN-based sparsity control (ablation) | UKPGAN · KeypointNet airplane, chair, and table | Rotation Rep. (%): 99.8 -> 98.3; 95.9 -> 88.3; 99.6 -> 90.6 | Hurt |
| add salient information distillation instead of averaging (ablation) | UKPGAN · KeypointNet airplane, chair, and table | IoU (%): 51.8 -> 68.8; 33.3 -> 36.2; 19.0 -> 34.7 | Improved |
| add salient information distillation instead of averaging (ablation) | UKPGAN · KeypointNet airplane, chair, and table | Rotation Rep. (%): 94.1 -> 98.3; 87.5 -> 88.3; 99.9 -> 90.6 | Mixed |
| use local rotation-invariant features instead of raw XYZ features (ablation) | UKPGAN · KeypointNet airplane, chair, and table | IoU (%): 22.4 -> 68.8; 16.0 -> 36.2; 21.2 -> 34.7 | Improved |
| use local rotation-invariant features instead of raw XYZ features (ablation) | UKPGAN · KeypointNet airplane, chair, and table | Rotation Rep. (%): 15.4 -> 98.3; 4.9 -> 88.3; 0.7 -> 90.6 | Improved |
| add symmetric regularization (ablation) | UKPGAN · KeypointNet airplane, chair, and table | IoU (%): 54.9 -> 68.8; 20.0 -> 36.2; 22.3 -> 34.7 | Improved |
| add symmetric regularization (ablation) | UKPGAN · KeypointNet airplane, chair, and table | Rotation Rep. (%): 85.1 -> 98.3; 77.0 -> 88.3; 73.0 -> 90.6 | Improved |
| use UKPGAN keypoints with PerfectMatch rather than random keypoints | 3DMatch registration · 3DMatch, 100 keypoints | Registration Recall (%): 16.1 -> 37.6 | Improved |
| use UKPGAN keypoints with PerfectMatch rather than random keypoints | ETH registration · ETH, 2500 keypoints | Registration Recall (%): 50.1 -> 58.2 | Improved |
| use UKPGAN keypoints with D3Feat descriptors rather than the D3Feat detector | 3DMatch registration · 3DMatch, 100 keypoints | Registration Recall (%): 67.2 -> 47.4 | Hurt |
| use UKPGAN keypoints with D3Feat descriptors rather than the D3Feat detector | ETH registration · ETH, 2500 keypoints | Registration Recall (%): 29.2 -> 28.3 | Hurt |
Closest papers by meaning
- Discovery of Latent 3D Keypoints via End-to-end Geometric Reasoning (NeurIPS 2018 · similarity 0.87)
- Key-Grid: Unsupervised 3D Keypoints Detection using Grid Heatmap Features (NeurIPS 2024 · similarity 0.87)
- SelfGeo: Self-supervised and Geodesic-consistent Estimation of Keypoints on Deformable Shapes (ECCV 2024 · similarity 0.87)
- KeypointDETR: An End-to-End 3D Keypoint Detector (ECCV 2024 · similarity 0.87)
- Back to 3D: Few-Shot 3D Keypoint Detection with Back-Projected 2D Features (CVPR 2024 · similarity 0.86)
- KeyPose: Category-Level 6D Object Pose Estimation with Self-Adaptive Keypoints (AAAI 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.
