Research map / Point clouds / Registration and matching
PCAM: Product of Cross-Attention Matrices for Rigid Registration of Point Clouds
Anh-Quan Cao, Gilles Puy, Alexandre Boulch, Renaud Marlet
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 |
|---|---|---|---|
| use product of attention matrices instead of last attention matrix only (ablation) | PCAM-Soft · 3DMatch val, L=6 | recall (%): 76.6 -> 82.4 | Improved |
| use product of attention matrices instead of last attention matrix only (ablation) | PCAM-Sparse · 3DMatch val, L=6 | recall (%): 78.4 -> 81.6 | Improved |
| use product of attention matrices instead of last attention matrix only (ablation) | PCAM-Soft · 3DMatch val, L=2 | recall (%): 37.1 -> 52.3 | Improved |
| use product of attention matrices instead of last attention matrix only (ablation) | PCAM-Sparse · 3DMatch val, L=2 | recall (%): 54.1 -> 52.4 | No clear effect |
| exchange context between point clouds via intermediate cross-attention matrices (ablation) | PCAM-Soft · 3DMatch val, L=6 | recall (%): 64.4 -> 76.6 | Improved |
| exchange context between point clouds via intermediate cross-attention matrices (ablation) | PCAM-Sparse · 3DMatch val, L=6 | recall (%): 68.0 -> 78.4 | Improved |
| increase number of layers from 2 to 6 (ablation) | PCAM-Soft · 3DMatch val | recall (%): 52.3 -> 82.4 | Improved |
| increase number of layers from 2 to 6 (ablation) | PCAM-Sparse · 3DMatch val | recall (%): 52.4 -> 81.6 | Improved |
| add LGeoConf confidence geometric loss (ablation) | PCAM-Soft · 3DMatch val, L=2 | recall (%): 45.9 -> 52.3 | Improved |
| add LGeoConf confidence geometric loss (ablation) | PCAM-Sparse · 3DMatch val, L=2 | recall (%): 48.5 -> 52.4 | Improved |
| replace classification attention loss by geometric attention loss (ablation) | PCAM-Soft · 3DMatch val, L=2 | recall (%): 52.3 -> 19.4 | Hurt |
| add geometric attention loss to classification attention loss (ablation) | PCAM-Soft · 3DMatch val, L=2 | recall (%): 52.3 -> 53.7 | Mixed |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- PointNetLK — PointNetLK: Robust & Efficient Point Cloud Registration Using PointNet (CVPR 2019) · 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
- DeepPRO: Deep Partial Point Cloud Registration of Objects (ICCV 2021 · similarity 0.89)
- Deep Closest Point: Learning Representations for Point Cloud Registration (ICCV 2019 · similarity 0.89)
- PosDiffNet: Positional Neural Diffusion for Point Cloud Registration in a Large Field of View with Perturbations (AAAI 2024 · similarity 0.88)
- 3D Focusing-and-Matching Network for Multi-Instance Point Cloud Registration (NeurIPS 2024 · similarity 0.87)
- PointMC: Multi-instance Point Cloud Registration based on Maximal Cliques (ICML 2024 · similarity 0.87)
- PCDreamer: Point Cloud Completion Through Multi-view Diffusion Priors (CVPR 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.
