Research map / Point clouds / Registration and matching
Density-invariant Features for Distant Point Cloud Registration
Quan Liu, Hongzi Zhu, Yunsong Zhou, Hongyang Li, Shan Chang, Minyi Guo
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 |
|---|---|---|---|
| replace pair-wise contrastive learning with group-wise contrastive learning (GCL) | distant point cloud registration, KITTI · KITTI [40,50] | registration recall (%): 26.7 -> 67.6 | Improved |
| GCL on nuScenes mean over distance bins | distant point cloud registration, nuScenes · nuScenes mRR | mRR (%): 39.5 -> 71.5 | Improved |
| GCL on easy close-range pairs | KITTI [5,10] and [10,20] · KITTI easy bins | registration recall (%): Predator 96.8 vs GCL+Conv 96.1 on [10,20] | No clear effect |
| increase number of point clouds per positive group phi from 2 to 6 (ablation) | GCL+Conv · LoKITTI | registration recall (%): 46.3 -> 72.3 | Improved |
| increase phi beyond 6 (8, 10) (ablation) | GCL+Conv · LoKITTI | registration recall (%): 72.3 -> 69.7 (phi=10) | Hurt |
| combine Finest loss with Positive Variance loss instead of either alone (ablation) | GCL+Conv · LoKITTI | registration recall (%): 64.4 (PV) -> 72.3 (F+PV); 53.2 (F) -> 72.3 | Improved |
| block gradient on the finest feature (BF) in finest loss (ablation) | GCL+Conv · LoKITTI | registration recall (%): 72.1 (F+PP) -> 50.1 (BF+PP) | Hurt |
| different loss designs on moderate-difficulty KITTI [10,10] (ablation) | GCL+Conv · KITTI [10,10] | registration recall (%): 98.6 to 99.2 | No clear effect |
| scale up pair-wise training data by 20x for FCGF instead of using GCL (ablation) | FCGF · LoKITTI | registration recall (%): 24.0 -> 28.8 (vs GCL 72.3) | No clear effect |
| use non-i.i.d. correlated positive pairs (PCL) rather than group positives | feature extractor training · feature drift analysis | feature drift | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 3 of 3 on most shared results.
- CoFiNet — CoFiNet: Reliable Coarse-to-fine Correspondences for Robust PointCloud Registration (NeurIPS 2021) · better on 2 of 3 shared results
- GeoTransformer — GeoTransformer: Fast and Robust Point Cloud Registration With Geometric Transformer (TPAMI 2023) · better on 1 of 1 shared result
- SpinNet — SpinNet: Learning a General Surface Descriptor for 3D Point Cloud Registration (CVPR 2021) · better on 1 of 1 shared result
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
- Sparse-to-Dense Matching Network for Large-Scale LiDAR Point Cloud Registration (TPAMI 2023 · similarity 0.89)
- HRegNet: A Hierarchical Network for Large-Scale Outdoor LiDAR Point Cloud Registration (ICCV 2021 · similarity 0.89)
- Mining and Transferring Feature-Geometry Coherence for Unsupervised Point Cloud Registration (NeurIPS 2024 · similarity 0.88)
- Unlocking Generalization Power in LiDAR Point Cloud Registration (CVPR 2025 · similarity 0.88)
- Extend Your Own Correspondences: Unsupervised Distant Point Cloud Registration by Progressive Distance Extension (CVPR 2024 · similarity 0.88)
- Cross-PCR: A Robust Cross-Source Point Cloud Registration Framework (AAAI 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.
