Research map / Point clouds / Registration and matching
GeoTransformer: Fast and Robust Point Cloud Registration With Geometric Transformer
Zheng Qin, Hao Yu, Changjian Wang, Yulan Guo, Yuxing Peng, Slobodan Ilić, Dewen Hu, Kai Xu
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 geometric self-attention instead of vanilla self-attention (ablation) | GeoTransformer · 3DLoMatch | Registration Recall (%): 68.4 -> 74.5 | Improved |
| add geometric self-attention instead of vanilla self-attention (ablation) | GeoTransformer · 3DLoMatch | Inlier Ratio (%): 32.6 -> 43.8 | Improved |
| use geometric self-attention instead of graph neural network (ablation) | GeoTransformer · 3DLoMatch | Registration Recall (%): 69.8 -> 74.5 | Improved |
| use overlap-aware circle loss instead of circle loss (ablation) | GeoTransformer · 3DLoMatch | Registration Recall (%): 71.5 -> 74.5 | Improved |
| use overlap-aware circle loss instead of cross-entropy loss (ablation) | GeoTransformer · 3DLoMatch | Registration Recall (%): 68.4 -> 74.5 | Improved |
| add triplet-wise angular embedding (k=3) to distance embedding (ablation) | GeoTransformer · 3DLoMatch | Registration Recall (%): 72.1 -> 74.5 | Improved |
| increase number of angle neighbors k from 3 to 4 (ablation) | GeoTransformer · 3DLoMatch | Registration Recall (%): 74.5 -> 75.1 | Improved |
| vary sigma_d from 0.2m to 0.5m (ablation) | GeoTransformer · 3DLoMatch | Registration Recall (%): 74.5 -> 73.7 | No clear effect |
| use max pooling instead of average pooling for angular embedding (ablation) | GeoTransformer · 3DLoMatch | Registration Recall (%): 74.0 -> 74.5 | Improved |
| add dual-normalization (ablation) | GeoTransformer · 3DLoMatch | Patch Inlier Ratio (%): 53.5 -> 54.6 | Improved |
| use LGR instead of weighted SVD (ablation) | GeoTransformer · 3DLoMatch | Registration Recall (%): 60.5 -> 74.5 | Improved |
| use LGR instead of RANSAC-50k for pose | GeoTransformer · 3DLoMatch | Registration Recall (%): 75.4 -> 74.5 | No clear effect |
Later papers that compared with this method
15 later accepted papers report GeoTransformer as a baseline in their results tables; 13 of them report a better number on most of the results they share with it.
- C-GenReg: Training-Free 3D Point Cloud Registration by Multi-View-Consistent Geometry-to-Image Generation with Probabilistic Modalities Fusion (CVPR 2026 · better on 7 of 8 shared results)
- Scalable and Differentiable Point-Cloud Registration Using Maximum Mean Discrepancy (ICML 2026 · better on 1 of 1 shared result)
- Generalized-CVO: Fast and Correspondence-Free Local Point Cloud Registration with Second Order Riemannian Optimization (CVPR 2026 · better on 1 of 1 shared result)
- Cross-PCR: A Robust Cross-Source Point Cloud Registration Framework (AAAI 2025 · better on 4 of 4 shared results)
- Rectified Point Flow: Generic Point Cloud Pose Estimation (NeurIPS 2025 · better on 2 of 2 shared results)
- RARE: Refine Any Registration of Pairwise Point Clouds via Zero-Shot Learning (ICCV 2025 · better on 2 of 2 shared results)
- PSReg: Prior-guided Sparse Mixture of Experts for Point Cloud Registration (AAAI 2025 · better on 1 of 1 shared result)
- UMERegRobust – Universal Manifold Embedding Compatible Features for Robust Point Cloud Registration (ECCV 2024 · better on 7 of 7 shared results)
7 more not listed.
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- CoFiNet — CoFiNet: Reliable Coarse-to-fine Correspondences for Robust PointCloud Registration (NeurIPS 2021) · better on 8 of 8 shared results
- Lepard — Lepard: Learning Partial Point Cloud Matching in Rigid and Deformable Scenes (CVPR 2022) · 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
- Geometric Transformer for Fast and Robust Point Cloud Registration (CVPR 2022 · similarity 0.97)
- RegFormer: An Efficient Projection-Aware Transformer Network for Large-Scale Point Cloud Registration (ICCV 2023 · similarity 0.90)
- Rethinking Point Cloud Registration as Masking and Reconstruction (ICCV 2023 · similarity 0.90)
- Efficient and Robust Point Cloud Registration via Heuristics-Guided Parameter Search (TPAMI 2024 · similarity 0.90)
- A Consistency-Aware Spot-Guided Transformer for Versatile and Hierarchical Point Cloud Registration (NeurIPS 2024 · similarity 0.89)
- Dual Focus-Attention Transformer for Robust Point Cloud Registration (CVPR 2025 · similarity 0.89)
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.
