Research map / Point clouds / Registration and matching
Learned feature-based registration
Register point clouds with learned features, multi-view and image cues, and unsupervised training.
66 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2017–2026) · an approach in Registration and matching, keyword cluster “cloud registration · overlap · feature”
State History Benchmarks Tried Baselines Checklist Preprints Who Export
Where it stands
Still moving The best reported result is still improving on 4 of 4 tracked benchmarks.
19 papers in 2023–24 and 29 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 54.3% to 63.0%.
Open questions the data points to
- 3DMatch (RR (%)): best went from 92 (up to 2024) to 95.8 in 2025 — Dual Focus-Attention Transformer for Robust Point Cloud Registration (CVPR 2025 · 15 citations)
- 3DLoMatch (RR (%)): best went from 64.8 (up to 2024) to 80.5 in 2025 — Dual Focus-Attention Transformer for Robust Point Cloud Registration (CVPR 2025 · 15 citations)
- ModelLoNet (RRE): best went from 3.578 (up to 2024) to 2.842 in 2025 — Partial Point Cloud Registration with Multi-view 2D Image Learning (AAAI 2025 · 1 citation)
- ModelNet (RRE): best went from 1.331 (up to 2024) to 1.138 in 2025 — Partial Point Cloud Registration with Multi-view 2D Image Learning (AAAI 2025 · 1 citation)
- Papers report Grid Cls. + PnP on KITTI (RRE (°)) anywhere from 10.15 to 19.19: comparisons against it depend on whose number you take.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2017 · 3D Point Cloud Registration for Localization Using a Deep Neural Network Auto-Encoder (CVPR 2017 · 255 citations)
- 2019 · PointNetLK: Robust & Efficient Point Cloud Registration Using PointNet (CVPR 2019 · 891 citations)
- 2020 · D3Feat: Joint Learning of Dense Detection and Description of 3D Local Features (CVPR 2020 · 519 citations)
- 2021 · Predator: Registration of 3D Point Clouds With Low Overlap (CVPR 2021 · 715 citations)
- 2022 · REGTR: End-to-End Point Cloud Correspondences With Transformers (CVPR 2022 · 272 citations)
- 2023 · 3D Registration With Maximal Cliques (CVPR 2023 · 208 citations)
- 2024 · FastMAC: Stochastic Spectral Sampling of Correspondence Graph (CVPR 2024 · 27 citations)
- 2025 · Dual Focus-Attention Transformer for Robust Point Cloud Registration (CVPR 2025 · 15 citations)
- 2026 · Adaptive Agent Selection and Interaction Network for Image-to-Point Cloud Registration (AAAI 2026 · 3 citations)
Earliest papers here
- 3D Point Cloud Registration for Localization Using a Deep Neural Network Auto-Encoder (CVPR 2017 · 255 citations)
- PointNetLK: Robust & Efficient Point Cloud Registration Using PointNet (CVPR 2019 · 891 citations)
- PRNet: Self-Supervised Learning for Partial-to-Partial Registration (NeurIPS 2019 · 112 citations)
Benchmark records
The best number each year that beat every earlier one, among papers of this approach reporting their own method. Numbers are extracted by AI from each paper's text; check the paper before citing.
3DMatch · RR (%) (higher is better)
6 papers, 2022–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 92 | REGTR: End-to-End Point Cloud Correspondences With Transformers (CVPR 2022 · 272 citations) |
| 2025 | 95.8 | Dual Focus-Attention Transformer for Robust Point Cloud Registration (CVPR 2025 · 15 citations) |
3DLoMatch · RR (%) (higher is better)
5 papers, 2022–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2022 | 64.8 | REGTR: End-to-End Point Cloud Correspondences With Transformers (CVPR 2022 · 272 citations) |
| 2025 | 80.5 | Dual Focus-Attention Transformer for Robust Point Cloud Registration (CVPR 2025 · 15 citations) |
ModelLoNet · RRE (lower is better)
4 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 5.235 | Predator: Registration of 3D Point Clouds With Low Overlap (CVPR 2021 · 715 citations) |
| 2022 | 3.93 | REGTR: End-to-End Point Cloud Correspondences With Transformers (CVPR 2022 · 272 citations) |
| 2023 | 3.578 | Unsupervised Deep Probabilistic Approach for Partial Point Cloud Registration (CVPR 2023 · 73 citations) |
| 2025 | 2.842 | Partial Point Cloud Registration with Multi-view 2D Image Learning (AAAI 2025 · 1 citation) |
ModelNet · RRE (lower is better)
4 papers, 2021–2025 · still improving
| Year | Best so far | Set by |
|---|---|---|
| 2021 | 1.739 | Predator: Registration of 3D Point Clouds With Low Overlap (CVPR 2021 · 715 citations) |
| 2022 | 1.473 | REGTR: End-to-End Point Cloud Correspondences With Transformers (CVPR 2022 · 272 citations) |
| 2023 | 1.331 | Unsupervised Deep Probabilistic Approach for Partial Point Cloud Registration (CVPR 2023 · 73 citations) |
| 2025 | 1.138 | Partial Point Cloud Registration with Multi-view 2D Image Learning (AAAI 2025 · 1 citation) |
What has been tried
48 papers here report 506 measured changes (a component added, removed or swapped, compared with a baseline): helped 73%, no clear effect 5%, hurt 20%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| Replace high-pass graph filtering with all-pass filtering | KITTI, FCGF · RR (%) | 71.89 -> 28.11 | Hurt | FastMAC: Stochastic Spectral Sampling of Correspondence Graph (CVPR 2024 · 27 citations) |
| Use skeleton-aware self-attention and vanilla cross-attention rather than both vanilla attention types | GermanyForest3D · RR (%) | 99.3 -> 97.4 | Hurt | SPEAL: Skeletal Prior Embedded Attention Learning for Cross-Source Point Cloud Registration (AAAI 2024 · 22 citations) |
| Remove Lowe filtering | KITTI · labeler IR | 5.1 -> 18.4; 7.8 -> 43.3 | Helped | Extend Your Own Correspondences: Unsupervised Distant Point Cloud Registration by Progressive Distance Extension (CVPR 2024 · 22 citations) |
| add instance-aware point matching | ROBI · MF (%) | 21.88 -> 39.80 | Helped | Learning Instance-Aware Correspondences for Robust Multi-Instance Point Cloud Registration in Cluttered Scenes (CVPR 2024 · 18 citations) |
| Use hue in geo-color positional embedding instead of geo-only embedding | Color3DMatch · Registration Recall (%) | 96.2 -> 97.3 | Helped | ColorPCR: Color Point Cloud Registration with Multi-Stage Geometric-Color Fusion (CVPR 2024 · 16 citations) |
| add superpoint focus-attention | 3DMatch · RR (%) | 93.6 -> 94.9 | Helped | Dual Focus-Attention Transformer for Robust Point Cloud Registration (CVPR 2025 · 15 citations) |
| Add virtual-spherical representation | KITTI Odometry, non-synchronized random-frame · RR (%) | 90.07 -> 95.40 | Helped | GraphI2P: Image-to-Point Cloud Registration with Exploring Pattern of Correspondence via Graph Learning (CVPR 2025 · 8 citations) |
| Add the window module | KITTI · RTE MAE (cm) | 4.34 -> 3.97 | Helped | PosDiffNet: Positional Neural Diffusion for Point Cloud Registration in a Large Field of View with Perturbations (AAAI 2024 · 8 citations) |
| add the interaction stage without the uncertainty estimation layer | RGB-D Scenes V2 · Inlier Ratio | Helped | Bridge 2D-3D: Uncertainty-aware Hierarchical Registration Network with Domain Alignment (AAAI 2025 · 7 citations) | |
| eliminate cross-attention | KITTI cross-distance generalization · mRR | 49.7 -> 58.8; 28.2 -> 87.7; 25.4 -> 47.3 | Mixed | Unlocking Generalization Power in LiDAR Point Cloud Registration (CVPR 2025 · 6 citations) |
| Add P2SG fusion of KP-Conv geometric features to the baseline graph matcher | 3RScan · F1 | 62.9 -> 79.0 | Helped | SG-PGM: Partial Graph Matching Network with Semantic Geometric Fusion for 3D Scene Graph Alignment and Its Downstream Tasks (CVPR 2024 · 6 citations) |
| Reduce sampled points from 4096 to 1024 | Scan2CAD · MR (%) | 95.44% -> 83.76% | Hurt | 3D Focusing-and-Matching Network for Multi-Instance Point Cloud Registration (NeurIPS 2024 · 6 citations) |
Baseline numbers that disagree
The same baseline on the same benchmark and metric, reported more than 5% apart by different papers here. Training budgets, data splits and reimplementations differ, so state which number you compare with.
| Baseline | Benchmark | Metric | Reported | Papers |
|---|---|---|---|---|
| Grid Cls. + PnP | KITTI | RRE (°) | 10.15 – 19.19 | 2 papers, e.g. CVPR 2021 and CVPR 2025 |
What reviewers will expect
Datasets and baselines named in the 17 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- 3DMatch · 7 of 17 papers
- ModelNet40 · 3 of 17 papers
- nuScenes · 3 of 17 papers
- 7-Scenes · 2 of 17 papers
- KITTI · 2 of 17 papers
- KITTI Odometry · 2 of 17 papers
- ScanNet · 2 of 17 papers
Baselines
- GeoTransformer · 10 of 17 papers
- Predator · 5 of 17 papers
- FCGF · 5 of 17 papers
- CoFiNet · 4 of 17 papers
- RegTR · 4 of 17 papers
- PEAL · 4 of 17 papers
- RPMNet · 3 of 17 papers
- RoITr · 3 of 17 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 48 papers; from OpenAlex and the papers' affiliations).
- Nanyang Technological University · 5 papers
- Shanghai Jiao Tong University · 4 papers
- Tsinghua University · 4 papers
- Xiamen University · 3 papers
- Xidian University · 3 papers
- Northwestern Polytechnical University · 3 papers
- University of Science and Technology of China · 2 papers
- New York University · 2 papers
- Xi’an Jiaotong University · 2 papers
- National University of Defense Technology · 2 papers
Use it
Download BibTeX of 40 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Robust inlier selection for registration (29 papers)
Elsewhere in the topic
- Non-rigid point set registration (42 papers)
- Quantum and certifiable correspondence (8 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).