Research map / Point clouds / Point cloud classification
SO-Net: Self-Organizing Network for Point Cloud Analysis
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 |
|---|---|---|---|
| add autoencoder pre-training (ablation) | SO-Net · ModelNet10 classification | accuracy: 94.4 -> 94.9 | Helped |
| add autoencoder pre-training (ablation) | SO-Net · ModelNet40 classification | accuracy: 89.3 -> 89.4 | Helped |
| add autoencoder pre-training (ablation) | SO-Net · ShapeNetPart | mean IoU: 84.6 -> 84.9 | Helped |
| add a third hierarchical feature-extraction layer (ablation) | SO-Net · ModelNet40 classification | accuracy: 89.3 -> 90.8 | Helped |
| increase hierarchical layer count too far (ablation) | SO-Net · ModelNet classification | accuracy | Hurt |
| replace the second fully connected layer block with 2D convolutions and pooling (ablation) | SO-Net · classification | accuracy | Hurt |
| use SOM nodes only and omit point-cloud coordinates (ablation) | SO-Net · ModelNet40 classification | accuracy: 92.3 -> 75.7 | Hurt |
| use random shift or rotation as data augmentation (ablation) | SO-Net · classification | accuracy | No clear effect |
| test with 50% of points missing (ablation) | SO-Net · ModelNet classification | accuracy: 1.7% | Hurt |
| test with 75% of points missing (ablation) | SO-Net · ModelNet classification | accuracy: 14.2% | Hurt |
| add Gaussian noise with sigma=0.2 to SOM nodes at test time (ablation) | SO-Net · ModelNet40 | accuracy: 91.1% | Hurt |
Closest papers by meaning
- DeepLA-Net: Very Deep Local Aggregation Networks for Point Cloud Analysis (CVPR 2025 · similarity 0.88)
- SOE-Net: A Self-Attention and Orientation Encoding Network for Point Cloud Based Place Recognition (CVPR 2021 · similarity 0.88)
- Self-Supervised Deep Learning on Point Clouds by Reconstructing Space (NeurIPS 2019 · similarity 0.87)
- PointNeXt: Revisiting PointNet++ with Improved Training and Scaling Strategies (NeurIPS 2022 · similarity 0.87)
- Learning to Sample (CVPR 2019 · similarity 0.87)
- Self-Supervised Global-Local Structure Modeling for Point Cloud Domain Adaptation With Reliable Voted Pseudo Labels (CVPR 2022 · similarity 0.86)
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.