Research map / Point clouds / Point cloud classification
SK-Net: Deep Learning on Point Cloud via End-to-End Discovery of Spatial Keypoints
Weikun Wu, Yan Zhang, David Wang, Yunqi Lei
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 |
|---|---|---|---|
| Use concatenated detail and pattern features instead of either feature alone (ablation) | SK-Net · ModelNet10 | accuracy: 93.9\% (detail only), 93.3\% (pattern only) -> 95.8\% | Improved |
| Add separation loss to classification loss without close loss (ablation) | SK-Net · ModelNet10 | accuracy: 94.3\% -> 93.8\% | Hurt |
| Add close loss to classification loss without separation loss (ablation) | SK-Net · ModelNet10 | accuracy: 94.3\% -> 93.2\% | Hurt |
| Add both separation and close losses (ablation) | SK-Net · ModelNet10 | accuracy: 94.3\% -> 95.8\%; 1.5\% | Improved |
| Replace kNN with ball query (ablation) | SK-Net · ModelNet10 | accuracy: 95.8\% -> 94.3\% (r=0.1); 94.7\% (r=0.2) | Hurt |
| Replace kNN with sift query (ablation) | SK-Net · ModelNet10 | accuracy: 95.8\% -> 92.5\% (r=0.1); 92.7\% (r=0.2) | Hurt |
| Use task-jointly learned Skeypoints rather than FPS- or SOM-derived keypoints (ablation) | SK-Net · ModelNet10 | accuracy | Improved |
| Use SK-Net rather than the strongest reported baseline | SK-Net · ModelNet10 | class accuracy: 95.5 -> 96.2 | Improved |
| Use SK-Net rather than the strongest reported baseline | SK-Net · ModelNet40 | instance accuracy: 93.4 -> 92.7 | Hurt |
| Use SK-Net rather than Point2Sequence | SK-Net · ShapeNetPart | mean IoU: 85.2 -> 85.0 | Hurt |
| Use SK-Net rather than PointNet++ (msg) | SK-Net · ScanNet | per-point accuracy: 83.2\% -> 81.4\% | Hurt |
| Test with 50% of input points missing | SK-Net · ModelNet10 and ModelNet40 | accuracy: drops by 3.4\% and 4.5\%, respectively | Hurt |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 2 on most shared results.
- SO-Net — SO-Net: Self-Organizing Network for Point Cloud Analysis (CVPR 2018) · better on 2 of 4 shared results
- Point2Sequence — Point2Sequence: Learning the Shape Representation of 3D Point Clouds with an Attention-Based Sequence to Sequence Network (AAAI 2019) · better on 0 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
- Mining Point Cloud Local Structures by Kernel Correlation and Graph Pooling (CVPR 2018 · similarity 0.88)
- Revisiting Point Cloud Shape Classification with a Simple and Effective Baseline (ICML 2021 · similarity 0.88)
- DeepLA-Net: Very Deep Local Aggregation Networks for Point Cloud Analysis (CVPR 2025 · similarity 0.88)
- PointAttN: You Only Need Attention for Point Cloud Completion (AAAI 2024 · similarity 0.87)
- PointCFormer: A Relation-Based Progressive Feature Extraction Network for Point Cloud Completion (AAAI 2025 · similarity 0.87)
- LinNet: Linear Network for Efficient Point Cloud Representation Learning (NeurIPS 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.
