Functional API with operators.

These are the building blocks for the layers. See The layer API for an easy to use high level interface.


build_spatial_hash_table(points, radius, …)

Creates a spatial hash table meant as input for fixed_radius_search

continuous_conv(filters, out_positions, …)

Continuous convolution of two pointclouds.

continuous_conv_backprop_filter(filters, …)

Computes the backprop for the filter of the ContinuousConv

continuous_conv_transpose(filters, …[, …])

Continuous tranpose convolution of two pointclouds.


Computes the backrop for the filter of the ContinuousConvTranspose

fixed_radius_search(points, queries, radius, …)

Computes the indices of all neighbors within a radius.

invert_neighbors_list(num_points, …[, name])

Inverts a neighbors list made of neighbors_index and neighbors_row_splits.

knn_search(points, queries, k, …[, …])

Computes the indices of k nearest neighbors.

nms(boxes, scores, nms_overlap_thresh[, name])

Performs non-maximum suppression of bounding boxes.

radius_search(points, queries, radii, …[, …])

Computes the indices and distances of all neighbours within a radius.

reduce_subarrays_sum(values, row_splits[, name])

Computes the sum for each subarray in a flat vector of arrays.

voxel_pooling(positions, features, voxel_size)

Spatial pooling for point clouds by combining points that fall into the same voxel bin.

voxel_pooling_grad(positions, features, …)

Gradient for features in VoxelPooling.

voxelize(points, row_splits, voxel_size, …)

Voxelization for point clouds.