Research map / Point clouds / Registration and matching
REGTR: End-to-End Point Cloud Correspondences With Transformers
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| directly predict correspondences with transformer attention instead of feature matching and RANSAC | REGTR · 3DMatch | RR (%): 90.5 -> 92.0 | Helped |
| directly predict correspondences with transformer attention instead of feature matching and RANSAC | REGTR · 3DLoMatch | RR (%): 62.5 -> 64.8 | Helped |
| add RANSAC after REGTR's predicted correspondences (ablation) | REGTR · 3DMatch | RR (%): 92.0 -> 91.9 | Hurt |
| add RANSAC after REGTR's predicted correspondences (ablation) | REGTR · 3DLoMatch | RR (%): 64.8 -> 63.3 | Hurt |
| decode correspondences as weighted coordinate sums rather than MLP regression (ablation) | REGTR · 3DMatch | RR (%): 92.0 -> 90.9 | Hurt |
| remove the feature loss (ablation) | REGTR · 3DMatch | RR (%): 92.0 -> 90.4 | Hurt |
| remove the feature loss (ablation) | REGTR · 3DLoMatch | RR (%): 64.8 -> 61.9 | Hurt |
| replace the feature loss with circle loss (ablation) | REGTR · 3DMatch and 3DLoMatch | RR (%): 3DMatch: 92.0 -> 90.0; 3DLoMatch: 64.8 -> 61.2 | Hurt |
| apply losses to all transformer layers instead of only the final layer (ablation) | REGTR · 3DMatch and 3DLoMatch | RR (%): 3DMatch: 92.0 -> 83.9; 3DLoMatch: 64.8 -> 46.8 | Hurt |
| use sinusoidal positional encoding rather than no positional encoding (ablation) | REGTR with weighted-coordinate decoding · 3DMatch | RR (%): 87.9 -> 90.9 | Helped |
| use sinusoidal positional encoding rather than no positional encoding (ablation) | REGTR with weighted-coordinate decoding · 3DLoMatch | RR (%): 55.0 -> 63.1 | Helped |
| replace learned positional encoding with sinusoidal encoding (ablation) | REGTR with weighted-coordinate decoding · 3DMatch and 3DLoMatch | RR (%): 3DMatch: 89.0 -> 90.9; 3DLoMatch: 61.3 -> 63.1 | Helped |
Closest papers by meaning
- Rectified Point Flow: Generic Point Cloud Pose Estimation (NeurIPS 2025 · similarity 0.88)
- Point Transformer V3: Simpler Faster Stronger (CVPR 2024 · similarity 0.88)
- Dual Focus-Attention Transformer for Robust Point Cloud Registration (CVPR 2025 · similarity 0.88)
- FreeReg: Image-to-Point Cloud Registration Leveraging Pretrained Diffusion Models and Monocular Depth Estimators (ICLR 2024 · similarity 0.88)
- Implicit Correspondence Learning for Image-to-Point Cloud Registration (CVPR 2025 · similarity 0.88)
- APML: Adaptive Probabilistic Matching Loss for Robust 3D Point Cloud Reconstruction (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.