class'L2', ignore_query_point=False, return_distances=False, normalize_distances=False, index_dtype=torch.int32, **kwargs)#

Radius search for 3D point clouds.

This layer computes the neighbors for each query point with each query having an individual radius.


This example shows a neighbor search that returns the indices to the found neighbors and the distances.:

import torch
import as ml3d

points = torch.randn([20,3])
queries = torch.randn([10,3])
radii = torch.randn([10])+1.0

nsearch = ml3d.layers.RadiusSearch(return_distances=True)
ans = nsearch(points, queries, radii)
# returns a tuple of neighbors_index, neighbors_row_splits, and neighbors_distance
  • metric – Either L1, L2 or Linf. Default is L2.

  • ignore_query_point – If True the points that coincide with the center of the search window will be ignored. This excludes the query point if ‘queries’ and ‘points’ are the same point cloud.

  • return_distances – If True the distances for each neighbor will be returned. If False a zero length Tensor will be returned instead.

  • normalize_distances – If True the returned distances will be normalized with the radii.

__init__(metric='L2', ignore_query_point=False, return_distances=False, normalize_distances=False, index_dtype=torch.int32, **kwargs)#

Initializes internal Module state, shared by both nn.Module and ScriptModule.

forward(points, queries, radii, points_row_splits=None, queries_row_splits=None)#

This function computes the neighbors within a radius for each query point.

  • points – The 3D positions of the input points.

  • queries – The 3D positions of the query points.

  • radii – A radius for each query point.

  • points_row_splits – Optional 1D vector with the row splits information if points is batched. This vector is [0, num_points] if there is only 1 batch item.

  • queries_row_splits – Optional 1D vector with the row splits information if queries is batched. This vector is [0, num_queries] if there is only 1 batch item.


3 Tensors in the following order


The compact list of indices of the neighbors. The corresponding query point can be inferred from the ‘neighbor_count_row_splits’ vector.


The exclusive prefix sum of the neighbor count for the query points including the total neighbor count as the last element. The size of this array is the number of queries + 1.


Stores the distance to each neighbor if ‘return_distances’ is True. Note that the distances are squared if metric is L2. This is a zero length Tensor if ‘return_distances’ is False.