yangyanli/PointCNN

Health v2 · maintenance only

Flags: no_license

Adoption not part of the score

1434 stars · 358 forks observed · 2026-08-28

What it is AI-extracted, prompt v1, taxonomy v1, 2026-08-30, confidence not recorded

PointCNN is a deep learning framework for feature learning from 3D point clouds, applying convolution on X-transformed points to handle their irregular, unordered structure. It achieved state-of-the-art results on benchmarks like ModelNet40, ScanNet, and ShapeNet Parts for classification and segmentation tasks.

Use cases

• classify 3D point clouds on ModelNet40
• segment parts of 3D shapes from ShapeNet
• semantic segmentation of indoor scenes from point clouds
• label voxels in ScanNet scans
• process LiDAR point clouds for autonomous driving
• segment aerial point clouds for 3D city mapping
• learn features from unordered 3D point sets

When to choose

• you need a proven point cloud CNN for classification or segmentation benchmarks
• you want pretrained models for point cloud tasks
• you're doing research on 3D feature learning from unordered point sets
• you need point cloud segmentation integrated into GIS workflows like ArcGIS

When to avoid

• you want the latest architecture - the authors recommend PointCNN++ instead
• you need a permissively licensed library - the license is non-standard
• you need active development or modern framework support
• you want a PyTorch-native implementation - use the PyTorch Geometric port instead

Facets

library · maturity maintenance

machine-learning deep-learning computer-vision image-processing computer-vision machine-learning deep-learning autonomous-vehicles robotics python point-cloud 3d-deep-learning pointcloud-segmentation pointcloud-classification neurips-2018 x-transformation tensorflow linux gpu

Member repositories

For agents

markdown · JSON · MCP: product_card(name="yangyanli/PointCNN")