Research map / Point clouds / Registration and matching
HyperGCT: A Dynamic Hyper-GNN-Learned Geometric Constraint for 3D Registration
Xiyu Zhang, Jiayi Ma, Jianwei Guo, Wei Hu, Zhaoshuai Qi, Fei Hui, Jiaqi Yang, Yanning Zhang
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 |
|---|---|---|---|
| model correspondences with a dynamic hypergraph learned by Hyper-GNN | HyperGCT · 3DLoMatch (FCGF) | registration recall (RR): 60.08 -> 63.73 | Improved |
| use HyperGCT instead of MAC on 3DMatch | HyperGCT · 3DMatch (FPFH) | registration recall (RR): 83.92 -> 85.89 | Improved |
| use HyperGCT instead of MAC on KITTI-LC 0-10m | HyperGCT · KITTI-LC 0-10m (FPFH) | registration recall (RR): 97.7 -> 97.05 | Hurt |
| use HyperGCT on high outlier KITTI-LC 20-30m | HyperGCT · KITTI-LC 20-30m (FPFH) | registration recall (RR): 23.73 -> 25.95 | Improved |
| train on KITTI-10m and test on 3DMatch (FPFH) | HyperGCT · KITTI-10m to 3DMatch generalization (FPFH) | registration recall (RR): 83.92 -> 85.40 | Improved |
| reduce compatibility threshold theta_cmp (HyperGCT vs MAC) (ablation) | HyperGCT · 3DLoMatch (FCGF) | registration recall (RR): MAC down 5.84% vs HyperGCT variation under 1% | Improved |
| increase graph updates from 1 to 5 (ablation) | HyperGCT · 3DMatch (FPFH) | registration recall (RR): 84.47 -> 85.89 | Improved |
| increase graph updates beyond 5 (7 or 9) (ablation) | HyperGCT · 3DMatch (FPFH) | registration recall (RR): 85.89 -> 84.72 | Hurt |
| add HyperGCT incidence matrix to SC2-PCR back-end (ablation) | SC2-PCR · 3DLoMatch (FCGF) | registration recall (RR): 58.62 -> 63.17 | Improved |
| add HyperGCT to PLE back-end (ablation) | Hunter PLE · 3DLoMatch (FCGF) | registration recall (RR): 60.02 -> 60.08 | No clear effect |
| use HyperGCT hypotheses instead of RANSAC, SAC-COT or MAC (ablation) | HyperGCT · 3DMatch (FPFH) | correctness of hypotheses (%): 6.64 -> 62.66 | Improved |
| use GF-NMS instead of standard NMS for seed selection (ablation) | HyperGCT · 3DMatch (FPFH) | average number of inlier seeds: 64 -> 152.65 | Improved |
Closest papers by meaning
- Multi-scale Consistency for Robust 3D Registration via Hierarchical Sinkhorn Tree (NeurIPS 2024 · similarity 0.87)
- Hyper-PCN: Hypergraph-Based Point Cloud Completion via High-Order Correlation Modeling (CVPR 2026 · similarity 0.87)
- Deep Graph-Based Spatial Consistency for Robust Non-Rigid Point Cloud Registration (CVPR 2023 · similarity 0.87)
- SG-PGM: Partial Graph Matching Network with Semantic Geometric Fusion for 3D Scene Graph Alignment and Its Downstream Tasks (CVPR 2024 · similarity 0.85)
- Scalable 3D Registration via Truncated Entry-wise Absolute Residuals (CVPR 2024 · similarity 0.85)
- DGC-GNN: Leveraging Geometry and Color Cues for Visual Descriptor-Free 2D-3D Matching (CVPR 2024 · similarity 0.85)
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.
