Overview

Tensor operations on packed point clouds: neighbor search, voxelization, geometry, boxes, and serialization.

Modules:

• anchors –

  Axis-aligned anchor generation and target assignment for the anchor-based BEV detectors.

• box3d –

  Generic 3D oriented-box geometry shared by detection models and their evaluation.

• cluster –

  Neighbor search and grouping: kNN, FPS, radius queries, local grids, decimation, and kNN interpolation.

• density –

  Gaussian kernel density estimation over packed batches.

• geometry –

  Geometric operations: bounding boxes, rotations, point transforms, vertex normals, and spherical sampling.

• heatmap –

  Gaussian center-heatmap targets for center-based 3D detection heads.

• hilbert –

  Hilbert curve encoding and decoding between coordinates and curve indices.

• math –

  Elementwise math: division with a default for zero denominators.

• octree –

  Octree construction, interpolation, and upsampling helpers built on ocnn.

• serialization –

  Space-filling curve serialization of voxel coordinates using Z-order and Hilbert encodings.

• voxelization –

  Dense and sparse voxelization, FNV voxel hashing, the first point per voxel, and trilinear devoxelization.