Research map / Point clouds / Registration and matching
3DFeat-Net: Weakly Supervised Local 3D Features for Point Cloud Registration
Zi Jian Yew, 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 the learned attention keypoint detector instead of ISS (ablation) | 3DFeat-Net descriptor · Oxford RobotCar registration | success rate: 97.66% -> 98.10% | Improved |
| use the learned detector instead of random sampling (ablation) | 3DFeat-Net descriptor · Oxford RobotCar registration | success rate: 90.28% -> 98.10% | Improved |
| use the learned detector instead of Random Sphere Cover Set (ablation) | 3DFeat-Net descriptor · Oxford RobotCar registration | success rate: 92.64% -> 98.10% | Improved |
| use the learned detector instead of ISS (ablation) | 3DFeat-Net descriptor · Oxford RobotCar registration | avg RANSAC iterations: 7127 -> 2940 | Improved |
| use keypoint detector (ISS) instead of random sampling for 3DMatch (ablation) | 3DMatch · Oxford RobotCar registration | success rate: 54.64% -> 69.06% | Improved |
| plug the learned descriptor into handcrafted detector rather than FPFH descriptor (ablation) | ISS keypoints · Oxford RobotCar registration | success rate: 92.32% -> 97.66% | Improved |
| train PointNet++ without descriptor alignment in the weak setting (ablation) | PointNet++ descriptor · Oxford RobotCar descriptor matching | error at 95% recall: 36.84 -> 50.57 | Hurt |
| increase descriptor dimension above 32 (ablation) | 3DFeat-Net descriptor · Oxford RobotCar descriptor matching | error at 95% recall | No clear effect |
| apply learned model to KITTI without fine-tuning versus ISS+FPFH | 3DFeat-Net · KITTI odometry registration (no fine-tuning) | success rate: 58.95% -> 95.97% | Improved |
| compare with LORAX on ETH using single best match and no refinement | 3DFeat-Net · ETH dataset registration (best point clouds) | success rate: 83% -> 100% | Improved |
Closest papers by meaning
- D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features (CVPR 2020 · similarity 0.91)
- HRegNet: A Hierarchical Network for Efficient and Accurate Outdoor LiDAR Point Cloud Registration (TPAMI 2023 · similarity 0.90)
- HRegNet: A Hierarchical Network for Large-Scale Outdoor LiDAR Point Cloud Registration (ICCV 2021 · similarity 0.90)
- GraphI2P: Image-to-Point Cloud Registration with Exploring Pattern of Correspondence via Graph Learning (CVPR 2025 · similarity 0.88)
- A Consistency-Aware Spot-Guided Transformer for Versatile and Hierarchical Point Cloud Registration (NeurIPS 2024 · similarity 0.88)
- Mining and Transferring Feature-Geometry Coherence for Unsupervised Point Cloud Registration (NeurIPS 2024 · similarity 0.87)
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.
