Research map / Point clouds / Shape and surface completion
USIP: Unsupervised Stable Interest Point Detection From 3D Point Clouds
Jiaxin Li, Gim Hee Lee
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 USIP detector instead of 3DFeatNet detector | Our Desc. on KITTI registration · KITTI, 256 keypoints | registration failure rate (%): 2.61 -> 1.41 | Improved |
| use USIP detector instead of 3DFeatNet detector | FPFH on KITTI registration · KITTI, 256 keypoints | registration failure rate (%): 12.15 -> 8.37 | Improved |
| use USIP detector instead of 3DFeatNet detector | SHOT on KITTI registration · KITTI, 256 keypoints | registration failure rate (%): 11.76 -> 5.40 | Improved |
| use USIP detector instead of 3DFeatNet detector | Our Desc. on KITTI registration · KITTI, 256 keypoints | inlier ratio (%): 15.66 -> 32.20 | Improved |
| use USIP detector in optimal configuration instead of 3DFeat-Net | registration pipeline · KITTI | registration failure rate (%): 0.57 -> 0.24 | Improved |
| use USIP detector instead of 3DFeatNet detector for speed | keypoint extraction · KITTI 128 keypoints | time (s): 0.438 -> 0.011 | Improved |
| train descriptor with USIP keypoints and saliency uncertainty instead of random keypoints (ablation) | Our Desc. · KITTI registration | registration failure rate (%): 1.02 -> 0.24 | Improved |
| train descriptor with USIP keypoints and saliency uncertainty instead of random keypoints (ablation) | Our Desc. · Oxford registration | registration failure rate (%): 1.20 -> 0.93 | Improved |
| increase number of neighbors K in the FPN (larger receptive field) (ablation) | USIP FPN · ModelNet40 (M=64) | keypoint degeneracy: K=9 no degeneracy; K=24 principal axis; K=64 centroid | Hurt |
| use too small number of proposals M or too large K (ablation) | USIP · point cloud registration | registration failure rate (%) | Hurt |
| add point-to-point loss (lambda=6 vs lambda=0) (ablation) | USIP · repeatability, 128 keypoints | relative repeatability | No clear effect |
| use large lambda for point-to-point loss (ablation) | USIP · training convergence | convergence | Hurt |
Later papers that compared with this method
2 later accepted papers report USIP as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Unsupervised Learning of 3D Semantic Keypoints with Mutual Reconstruction (ECCV 2022 · better on 1 of 1 shared result)
- RSKDD-Net: Random Sample-based Keypoint Detector and Descriptor (NeurIPS 2020 · better on 8 of 11 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
- Not All Points Are Equal: Learning Highly Efficient Point-Based Detectors for 3D LiDAR Point Clouds (CVPR 2022 · similarity 0.87)
- DH3D: Deep Hierarchical 3D Descriptors for Robust Large-Scale 6DoF Relocalization (ECCV 2020 · similarity 0.87)
- D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features (CVPR 2020 · similarity 0.86)
- Learning Transferable Features for Point Cloud Detection via 3D Contrastive Co-training (NeurIPS 2021 · similarity 0.86)
- GBlobs: Explicit Local Structure via Gaussian Blobs for Improved Cross-Domain LiDAR-based 3D Object Detection (CVPR 2025 · similarity 0.86)
- LiDAR-Net: A Real-scanned 3D Point Cloud Dataset for Indoor Scenes (CVPR 2024 · 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.
