Research map / Point clouds / Point cloud classification
P2-Net: Joint Description and Detection of Local Features for Pixel and Point Matching
Bing Wang, Changhao Chen, Zhaopeng Cui, Jie Qin, Chris Xiaoxuan Lu, Zhengdi Yu, Peijun Zhao, Zhen Dong, Fan Zhu, Niki Trigoni, Andrew Markham
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 |
|---|---|---|---|
| add jointly learned detector (vs no detector, random keypoints) (ablation) | P2-Net · 7Scenes pixel-point matching (Stairs) | Feature Matching Recall (%): 15.0 -> 88.8 | Improved |
| use proposed batch-hard detector instead of SIFT keypoints (ablation) | P2-Net · 7Scenes pixel-point matching (Stairs) | Feature Matching Recall (%): 15.8 -> 88.8 | Improved |
| use proposed detector instead of D2-Net detector (ablation) | P2-Net · 7Scenes pixel-point matching (Stairs) | Feature Matching Recall (%): 74.3 -> 88.8 | Improved |
| use proposed detector instead of D3Feat detector (ablation) | P2-Net · 7Scenes pixel-point matching (Stairs) | Feature Matching Recall (%): 17.8 -> 88.8 | Improved |
| use detector-guided keypoints at test instead of random sampling (ablation) | P2-Net · 7Scenes pixel-point matching (Stairs) | Registration Recall (%): 74.3 -> 82.6 | Improved |
| add ultra-wide reception 2D extractor (vs R2D2 network) (ablation) | P2-Net · 7Scenes pixel-point matching (Stairs) | Feature Matching Recall (%): 16.2 -> 88.8 | Improved |
| add ultra-wide reception 2D extractor (vs R2D2 network) (ablation) | P2-Net · 7Scenes pixel-point matching (Stairs) | Keypoint Repeatability (%): 33.1 -> 48.1 | Improved |
| add ultra-wide reception 2D extractor (vs ASL network) (ablation) | P2-Net · 7Scenes pixel-point matching (Stairs) | Registration Recall (%): 38.2 -> 82.6 | Improved |
| use circle-guided descriptor loss instead of D2-Net triplet loss | D2-Net · HPatches (viewpoint) | Recall @3px: 0.382 -> 0.413 | Improved |
| use P2-Loss instead of contrastive loss | ASLFeat · HPatches (viewpoint) | Recall @3px: 0.583 -> 0.625 | Improved |
| use P2-Loss instead of D2-Net triplet loss under illumination change | D2-Net · HPatches (illumination) | Recall @3px: 0.564 -> 0.560 | No clear effect |
| use P2-Loss instead of contrastive loss under illumination change | ASLFeat · HPatches (illumination) | HEstimation: 0.919 -> 0.915 | No clear effect |
Closest papers by meaning
- D2-Net: A Trainable CNN for Joint Description and Detection of Local Features (CVPR 2019 · similarity 0.88)
- EP2P-Loc: End-to-End 3D Point to 2D Pixel Localization for Large-Scale Visual Localization (ICCV 2023 · similarity 0.88)
- EdgePoint2: Compact Descriptors for Superior Efficiency and Accuracy (IJCV 2026 · similarity 0.88)
- Image2Point: 3D Point-Cloud Understanding with 2D Image Pretrained Models (ECCV 2022 · similarity 0.88)
- DGC-GNN: Leveraging Geometry and Color Cues for Visual Descriptor-Free 2D-3D Matching (CVPR 2024 · similarity 0.87)
- Rooms from Motion: Un-posed Indoor 3D Object Detection as Localization and Mapping (NeurIPS 2025 · 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.
