Tutorials

Four notebooks work through the Stage-2 workflow end to end on a small, real LiDAR sample — a ~150 m crop of the Dutch national dataset AHN4 (public domain, CC0), included in the repository under examples/data/. Everything runs on a CPU; no GPU and no training data are required.

Install the feature extras first:

pip install "geoai3d[laz,viz,gis]"

The notebooks live in examples/notebooks/ and render directly on GitHub:

1. Getting started — read a tile, inspect its CRS and classes, reproject, compute geometric features, and view the cloud in 3D.
2. Terrain and volumes — filter the ground, build a DTM and DSM, difference them for object heights, compute a volume, and export GeoTIFFs.
3. Unsupervised segmentation — decompose the scene into ground and objects with no labels.
4. Classification and GIS export — engineer features, train a random forest, evaluate it with spatially-blocked cross-validation, and export a classified GeoPackage.

To run them without Jupyter, the same pipelines are collected in a plain script you can execute directly:

python examples/verify_notebooks.py