Research map / Point clouds / Registration and matching
Learning Multiview 3D Point Cloud Registration
Zan Gojcic, Caifa Zhou, Jan D. Wegner, Leonidas J. Guibas, Tolga Birdal
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 |
|---|---|---|---|
| Use the learned pairwise registration network instead of descriptor matching with RANSAC | 3DMatch pairwise registration · 3DMatch | registration recall: 0.86 -> 0.87 | Improved |
| Add iterative refinement using previous weights and residuals (ablation) | Ours pairwise registration · 3DMatch | registration recall: 0.87 -> 0.89 | Improved |
| Synchronize pairwise estimates into global poses | Ours multiview registration · ScanNet (Ours, Good) | rotation ECDF at 3°: 57.7 -> 65.8 | Improved |
| Synchronize pairwise estimates into global poses | Ours multiview registration · ScanNet (Ours, Good) | translation ECDF at 0.05 m: 44.8 -> 48.4 | Improved |
| Use confidence-estimation edge pruning rather than no edge pruning (ablation) | Ours multiview registration · ScanNet | registration performance | Improved |
| Use local and global confidence combined by harmonic mean rather than relying only on local confidence (ablation) | Ours transformation synchronization · ScanNet | registration performance | Improved |
| Prune low-confidence edges before transformation synchronization (ablation) | Ours multiview registration · ScanNet | registration performance | Improved |
| Feed iterative weights and residuals back when outliers are prevalent or edge pruning is ineffective (ablation) | Ours pairwise registration · ScanNet | registration performance | Hurt |
| Increase the edge-pruning threshold above 0.2 (ablation) | Ours multiview registration · ScanNet | angular and translation errors | No clear effect |
| Replace RANSAC with the learned softNN pairwise registration | 3DMatch Kitchen scene · 3DMatch Kitchen scene (60 fragments) | total runtime (s): 1106.3 -> 80.3 | Improved |
Closest papers by meaning
- Register Any Point: Scaling 3D Point Cloud Registration by Flow Matching (ECCV 2026 · similarity 0.92)
- Robust Multiview Point Cloud Registration With Reliable Pose Graph Initialization and History Reweighting (CVPR 2023 · similarity 0.92)
- Partial Point Cloud Registration with Multi-view 2D Image Learning (AAAI 2025 · similarity 0.91)
- 3D Focusing-and-Matching Network for Multi-Instance Point Cloud Registration (NeurIPS 2024 · similarity 0.90)
- PointMC: Multi-instance Point Cloud Registration based on Maximal Cliques (ICML 2024 · similarity 0.89)
- Multiway Point Cloud Mosaicking with Diffusion and Global Optimization (CVPR 2024 · similarity 0.88)
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.
