Research map / Point clouds / Segmentation and detection
Pointwise Convolutional Neural Networks
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 pointwise convolution for point-cloud segmentation | Pointwise CNN · S3DIS | accuracy: 81.5 vs 87.0 | Mixed |
| Use pointwise convolution for point-cloud object recognition | Pointwise CNN · ModelNet40 | accuracy: 86.1 vs 90.7 | Hurt |
| Use pointwise convolution for object recognition on reconstructed and CAD objects | Pointwise CNN · ObjectNN | accuracy: 65.1 vs 65.6 | No clear effect |
| Add feature concatenation (ablation) | Pointwise CNN · ModelNet40 | accuracy: 78.6 -> 78.0 | Hurt |
| Add à-trous convolution (ablation) | Pointwise CNN · ModelNet40 | accuracy: 78.6 -> 75.0 | Hurt |
| Combine feature concatenation and à-trous convolution (ablation) | Pointwise CNN · ModelNet40 | accuracy: 78.6 -> 82.5 | Helped |
| Add SELU activation (ablation) | Pointwise CNN · ModelNet40 | accuracy: 78.6 -> 81.7 | Helped |
| Combine feature concatenation and SELU activation (ablation) | Pointwise CNN · ModelNet40 | accuracy: 78.6 -> 81.9 | Helped |
| Combine feature concatenation, à-trous convolution, and SELU (ablation) | Pointwise CNN · ModelNet40 | accuracy: 78.6 -> 85.2 | Helped |
| Add dropout to the full feature-concatenation, à-trous, and SELU configuration (ablation) | Pointwise CNN · ModelNet40 | accuracy: 85.2 -> 86.1 | Helped |
| Order points by Morton curve rather than ZYX (ablation) | Pointwise CNN · ModelNet40 | accuracy: 86.1 -> 86.0 | No clear effect |
| Replace fixed-radius neighbor query with k-nearest neighbors (ablation) | Pointwise CNN · ModelNet40 | accuracy: 86.1 -> 85.7 | No clear effect |
Closest papers by meaning
- PointNet: Deep Learning on Point Sets for 3D Classification and Segmentation (CVPR 2017 · similarity 0.92)
- A-CNN: Annularly Convolutional Neural Networks on Point Clouds (CVPR 2019 · similarity 0.91)
- PointConv: Deep Convolutional Networks on 3D Point Clouds (CVPR 2019 · similarity 0.91)
- Mining Point Cloud Local Structures by Kernel Correlation and Graph Pooling (CVPR 2018 · similarity 0.88)
- Shape-Oriented Convolution Neural Network for Point Cloud Analysis (AAAI 2020 · similarity 0.88)
- Attentional ShapeContextNet for Point Cloud Recognition (CVPR 2018 · similarity 0.88)
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.