Examples

The repository includes small notebook-style examples under examples/. They use tiny tensors and simple PyTorch losses to show where each geometry primitive fits in an unstructured point-cloud training step.

Each notebook is CUDA-gated. If CUDA is unavailable, the notebook prints a clear skip message instead of running the CUDA cells.

Notebooks

Notebook	Demonstrates
basic_bvh_knn.ipynb	local point-neighborhood feature aggregation with BVH(points) and bvh.knn(...)
mls_interpolation.ipynb	MLS feature sampling at learned offset positions, including return_grad=True field gradients
batched_displaced_query.ipynb	a tiny multihead displaced-query block with query_displaced_knn, gather_neighbor_values, and interpolate_displaced
fps_downsampling_geometry.ipynb	point-cloud downsampling with fps(...)