Research map / Point clouds / Registration and matching
PRNet: Self-Supervised Learning for Partial-to-Partial Registration
Read the paperCheck 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 |
|---|---|---|---|
| Use PRNet instead of FGR | Partial-to-partial registration on ModelNet40 test objects · ModelNet40 test objects | MAE(R): 2.832 -> 1.454 | Helped |
| Train PRNet on ShapeNetCore instead of ModelNet40 (ablation) | PRNet · ModelNet40 held-out categories | MAE(R): 2.329 -> 1.712 | Helped |
| Use predicted temperature λ rather than fixed, annealed, or learned λ (ablation) | PRNet · ModelNet40 partial-to-partial registration | MAE(R): Fixed λ: 1.659; Annealed λ: 1.630; Learned λ: 1.675; Predicted λ: 1.454 | Helped |
| Select keypoints by feature L2 norm instead of random sampling or closeness (ablation) | PRNet · ModelNet40 partial-to-partial registration | MAE(R): Random sampling: 1.689; Closeness: 2.109; L2 Norm: 1.454 | Helped |
| Increase the number of selected keypoints from 16 to 512 (ablation) | PRNet · ModelNet40 partial-to-partial registration | MAE(R): 27.843 -> 1.454 | Helped |
| Increase missing-data ratio from 25% to 75% (ablation) | PRNet · ModelNet40 partial-to-partial registration | MAE(R): 1.454 -> 6.447 | Hurt |
| Increase test noise from σ=0.01 to σ=0.1 (ablation) | PRNet · ModelNet40 partial-to-partial registration | MAE(R): 2.051 -> 5.013 | Hurt |
| Use discount factor 0.9 instead of 0.5, 0.7, or 0.99 (ablation) | PRNet · ModelNet40 partial-to-partial registration | MAE(R): 0.5: 1.921; 0.7: 1.998; 0.9: 1.454; 0.99: 1.732 | Helped |
| Train on registration representations and transfer them to a linear SVM | PRNet representations · ModelNet40 classification | accuracy: z/z: 85.2; SO(3)/SO(3): 80.5 | Mixed |
| Use PRNet instead of DCP-v2 | Partial-to-partial registration with Gaussian noise · ModelNet40 test objects with Gaussian noise | MAE(R): 4.534 -> 2.051 | Helped |
Closest papers by meaning
- Deep Confidence Guided Distance for 3D Partial Shape Registration (AAAI 2022 · similarity 0.90)
- PointNetLK: Robust & Efficient Point Cloud Registration Using PointNet (CVPR 2019 · similarity 0.89)
- Rectified Point Flow: Generic Point Cloud Pose Estimation (NeurIPS 2025 · similarity 0.86)
- Implicit Correspondence Learning for Image-to-Point Cloud Registration (CVPR 2025 · similarity 0.86)
- Mining and Transferring Feature-Geometry Coherence for Unsupervised Point Cloud Registration (NeurIPS 2024 · similarity 0.86)
- PSReg: Prior-guided Sparse Mixture of Experts for Point Cloud Registration (AAAI 2025 · 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.