Skip to content

Tutorials

Guided, runnable notebooks covering the library end to end. Each page is rendered from a Jupyter notebook committed under docs/examples/; open it in Colab or download it from the badges at the top of every tutorial.

The three tiers build on each other. Beginner gets a pretrained model running and explains the conventions everything else assumes. Intermediate puts your own data and your own training loop in the middle. Advanced takes the library to the shapes real projects ship in: survey tiles too large for one forward pass, driving sweeps, and whole rooms read by several models at once.

Beginner

Start here if you have not run a point cloud model before. Everything in this tier runs from committed sample data, with no dataset download.

Intermediate

Your data, your training loop, and the representation a trained model leaves behind.

  • Use your own data

    Several clouds collated into one packed batch, one color per sample

    Wrap your own point cloud files in a dataset with disk caching and packed-batch collation.

  • Train a model

    A training curve falling over successive epochs

    Train a classification model from scratch: dataloaders, optimizer, and the evaluation loop.

  • Features and similarity search

    A scene colored by the principal components of a pretrained encoder's features

    Read a frozen encoder's features, query one point, and retrieve whole shapes by descriptor.

Advanced

Production-shaped problems: clouds too large for one forward pass, oriented boxes in traffic, and several models combined into a single answer.