pypose.nbr_filter

class pypose.nbr_filter(points, nbr, radius, pdim=None, ord=2, return_mask=False)

Filter point outliers by checking if a point has less than n neighbors (nbr) within a radius.

Parameters:

• points (torch.Tensor) – the input point cloud. It is possible that the last dimension (D) is larger than pdim, with point’s coordinates using first pdim values. Subsequent values may contain additional information like intensity, RGB channels, etc. The shape has to be (N, D).

• nbr (int) – the minimum number of neighbors (nbr) within a certain radius.

• ord (int, optional) – the order of norm to use for distance calculation. Default: 2 (Euclidean distance).

• radius (float) – the radius of the sphere for counting the neighbors.

• pdim (int, optional) – the dimsion of points, where \(\text{pdim} \le D\). Default to the last dimension of points, if None.

• return_mask (bool, optional) – return the mask of inliers of not.

Returns:

The point clouds removed outliers. torch.BoolTensor: The mask of point clouds removed outliers, where the inlier is True and the outlier is False. The shape is (…, N).

Return type:

torch.Tensor

Warning

Note that this operation does not support batch operations, since the number of output voxels can be different on different batches.

Example

>>> import torch, pypose as pp
>>> points = torch.tensor([[0., 0., 0.],
...                        [1., 0., 0.],
...                        [0., 1., 0.],
...                        [0., 1., 1.],
...                        [10., 1., 1.],
...                        [10., 1., 10.]])
>>> pp.nbr_filter(points, nbr=2, radius=5)
tensor([[0., 0., 0.],
        [1., 0., 0.],
        [0., 1., 0.],
        [0., 1., 1.]])
>>> pp.nbr_filter(points, nbr=2, radius=12, return_mask=True)
(tensor([[ 0.,  0.,  0.],
         [ 1.,  0.,  0.],
         [ 0.,  1.,  0.],
         [ 0.,  1.,  1.],
         [10.,  1.,  1.]]),
tensor([ True,  True,  True,  True,  True, False]))