Research map / Point clouds / Point cloud classification
Global-Local Bidirectional Reasoning for Unsupervised Representation Learning of 3D Point Clouds
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 |
|---|---|---|---|
| replace supervised PointNet++ (Large) training with unsupervised GLR training | PointNet++ (Large) · ModelNet40 | accuracy (%): 92.01 -> 93.02 | Helped |
| replace supervised training with unsupervised GLR training | PointNet++ (Large) · ScanObjectNN | accuracy (%): 84.3 -> 87.2 | Helped |
| add GLR to supervised training | PointNet++ (Small) · ModelNet40 | accuracy (%): 91.69 -> 92.42 | Helped |
| add GLR to supervised training | PointNet++ (Large) · ModelNet40 | accuracy (%): 92.01 -> 92.76 | Hurt |
| use local-to-global reasoning instead of self-reconstruction only (ablation) | PointNet++ (Small) · ModelNet40 | accuracy (%): 86.77 -> 90.02 | Helped |
| add self-reconstruction to local-to-global reasoning (ablation) | PointNet++ (Small) · ModelNet40 | accuracy (%): 90.02 -> 90.96 | Helped |
| add multi-level feature aggregation (ablation) | PointNet++ (Small) · ModelNet40 | accuracy (%): 90.96 -> 91.69 | Helped |
| add normal estimation supervision (ablation) | PointNet++ (Small) · ModelNet40 | accuracy (%): 91.69 -> 92.22 | Helped |
| add ShapeNet training data (ablation) | full GLR PointNet++ (Small) · ModelNet40 | accuracy (%): 92.22 -> 92.30; 0.08% | Helped |
| increase the PointNet++ channel width | unsupervised GLR PointNet++ · ModelNet40 | accuracy (%): 92.22 -> 93.02 | Helped |
| train on the unsupervised representation rather than the supervised representation | PointNet++ (Large) representation · ModelNet40 -> ScanObjectNN | accuracy (%): 65.92 -> 87.22; +21.30 | Helped |
| train on the unsupervised representation rather than the supervised representation | PointNet++ (Large) representation · ModelNet40 -> ScanNet | accuracy (%): 77.31 -> 89.23; +11.92 | Helped |
Closest papers by meaning
- Learning Generalizable Part-based Feature Representation for 3D Point Clouds (NeurIPS 2022 · similarity 0.89)
- Mining Point Cloud Local Structures by Kernel Correlation and Graph Pooling (CVPR 2018 · similarity 0.89)
- Sonata: Self-Supervised Learning of Reliable Point Representations (CVPR 2025 · similarity 0.88)
- Geometrically-driven Aggregation for Zero-shot 3D Point Cloud Understanding (CVPR 2024 · similarity 0.88)
- Interpretable3D: An Ad-Hoc Interpretable Classifier for 3D Point Clouds (AAAI 2024 · similarity 0.87)
- Multi-View Representation is What You Need for Point-Cloud Pre-Training (ICLR 2024 · 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.