Research map / Point clouds / Registration and matching
Rethinking Point Cloud Registration as Masking and Reconstruction
Guangyan Chen, Meiling Wang, Li Yuan, Yi Yang, Yufeng Yue
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 the Masked Reconstruction Auxiliary Network (MRA) during training (ablation) | MRT · 3DLoMatch | RR (%): 67.8 -> 75.4 | Improved |
| add MRA to Predator | Predator · 3DLoMatch | RR (%): 62.5 -> 67.3 | Improved |
| add MRA to GeoTR | GeoTR · 3DLoMatch | RR (%): 75.0 -> 77.1 | Improved |
| add MRA to RegTR | RegTR · 3DLoMatch | RR (%): 64.8 -> 71.0 | Improved |
| additionally randomly mask point patches before the encoder (ablation) | MRT · 3DLoMatch | RR (%): 75.4 -> 73.6 | Hurt |
| use transformed center points for mask token positional encodings (ablation) | MRT · 3DLoMatch | RR (%): 71.6 -> 75.4 | Improved |
| also reconstruct overlapping regions of the other point cloud (ablation) | MRT · 3DLoMatch | RR (%): 73.2 -> 75.4 | Improved |
| use l2 Chamfer loss instead of l1 (ablation) | MRT · 3DLoMatch | RR (%): 72.4 -> 75.4 | Improved |
| use l2 Chamfer loss instead of l1+l2 (ablation) | MRT · 3DMatch | RR (%): 95.1 -> 95.1 | No clear effect |
| add Deviation Correction module (ablation) | MRT · 3DLoMatch | RR (%): 73.5 -> 75.4 | Improved |
| predict deviation compensation instead of directly projecting coordinates (ablation) | MRT · 3DLoMatch | RR (%): 71.0 -> 75.4 | Improved |
| add overlap prediction handled in full model vs without overlap reconstruction on 3DMatch RRE (ablation) | MRT · 3DMatch | RRE (deg): 1.453 -> 1.324 | Improved |
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.
- RegTR — REGTR: End-to-End Point Cloud Correspondences With Transformers (CVPR 2022) · better on 4 of 4 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 0.90)
- Robust Point Cloud Registration Framework Based on Deep Graph Matching (CVPR 2021 · similarity 0.90)
- GeoTransformer: Fast and Robust Point Cloud Registration With Geometric Transformer (TPAMI 2023 · similarity 0.90)
- Distinctiveness Oriented Positional Equilibrium for Point Cloud Registration (ICCV 2021 · similarity 0.90)
- 3D Focusing-and-Matching Network for Multi-Instance Point Cloud Registration (NeurIPS 2024 · similarity 0.90)
- A Consistency-Aware Spot-Guided Transformer for Versatile and Hierarchical Point Cloud Registration (NeurIPS 2024 · 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.
