Deep Learning for Point Clouds

This repository includes a pytorch implementation and training of a PointNet for 3D point cloud classification.

The included files are the following:

• PointNet Pytorch Implementation.ipynb: The PointNet implementation in pytorch and explains each step
• PointNet Training and Evaluation.ipynb: The training and evaluation of the PointNet with a subset of the ShapeNet dataset

Results

The trained model achieved 93% accuracy on the test set. Results from the test set can be checked in the following confusion matrix.

The model also shows the main points from the input point cloud used for the prediction.