v1.0.0 — Initial Release

PointNet-PyTorch v1.0.0

First stable release: a from-scratch PyTorch reimplementation of PointNet (Qi et al., 2017) for point cloud classification on ModelNet40.

Highlights

• Full PointNet architecture: input T-Net, feature T-Net, shared MLPs, symmetric max pooling, FC classifier head
• Modular PointNetEncoder backbone separated from the classification head, ready for future extensions (e.g. segmentation)
• Config-driven ablation flags (input/feature transform toggle, pooling type, regularization weight, data augmentation)
• 86.43% test accuracy on ModelNet40 (100 epochs, Colab T4 GPU)
• Full test suite covering T-Net, classifier, loss, and dataset components
• Per-class accuracy reporting + confusion matrix visualization

Known Limitations

• No segmentation head yet (encoder is designed to support one)
• No PointNet++ hierarchical feature learning
• Ablation experiment results not yet fully populated

Full Changelog: https://github.com/mariiammaysara/PointNet-PyTorch/commits/v1.0.0