filters

voxel.filters.gaussian_filter(volume, sigma, *components, space=None, truncate=2.0, stride=1, separable=True, padding_mode='replicate') -> vx.Volume

Apply Gaussian smoothing to a volume.

Parameters:

• sigma (float or Tensor) –

  Standard deviation(s) of size \((1,)\) or \((3,)\).

• *components (float, default: () ) –

  Additional components of sigma, allowing values to be passed as separate positional arguments, e.g. gaussian_filter(vol, 1, 1, 2, 'voxel').

• space (Space, default: None ) –

  The space of the sigma values, either 'voxel' or 'world'. Can be provided as the last positional argument.

• truncate (float, default: 2.0 ) –

  The number of standard deviations to extend the kernel before truncating.

• stride (int or Tensor, default: 1 ) –

  Downsampling stride(s) in voxel units.

• separable (bool, default: True ) –

  Whether to filter with three separable 1D kernels or a single dense 3D kernel. The results are equivalent.

• padding_mode (str, default: 'replicate' ) –

  Border padding mode, e.g. 'replicate' or 'zeros'.

Returns:

• Volume –

  Smoothed floating-point volume.

voxel.filters.box_filter(volume, size, *components, space=None, stride=1, separable=True, padding_mode='replicate') -> vx.Volume

Apply mean filtering to a volume with a box kernel.

The box extent is rounded to the nearest odd number of voxels along each dimension.

Parameters:

• size (float or Tensor) –

  Box extent(s) of size \((1,)\) or \((3,)\).

• *components (float, default: () ) –

  Additional components of size, allowing values to be passed as separate positional arguments, e.g. box_filter(vol, 3, 3, 1, 'voxel').

• space (Space, default: None ) –

  The space of the size values, either 'voxel' or 'world'. Can be provided as the last positional argument.

• stride (int or Tensor, default: 1 ) –

  Downsampling stride(s) in voxel units.

• separable (bool, default: True ) –

  Whether to filter with three separable 1D kernels or a single dense 3D kernel. The results are equivalent.

• padding_mode (str, default: 'replicate' ) –

  Border padding mode, e.g. 'replicate' or 'zeros'.

Returns:

• Volume –

  Filtered floating-point volume.

voxel.filters.apply_filter(volume, kernel, stride=1, padding_mode='replicate') -> vx.Volume

Apply a filter kernel to a volume with 'same'-style output extents.

Channels are filtered independently. When strided, the grid is downsampled to a spatial size of \(ceil(baseshape / stride)\) and the volume geometry is updated to reflect the new voxel spacing.

Parameters:

• kernel (Tensor or list) –

  A single dense 3D kernel of shape \((W, H, D)\) or a sequence of three 1D kernels applied separably per dimension.

• stride (int or Tensor, default: 1 ) –

  Downsampling stride(s) in voxel units.

• padding_mode (str, default: 'replicate' ) –

  Border padding mode, e.g. 'replicate' or 'zeros'.

Returns:

• Volume –

  Filtered floating-point volume.

voxel.filters.gaussian_kernel_1d(sigma, truncate=2.0, device=None, dtype=None) -> torch.Tensor

Generate a 1D Gaussian kernel with a specified standard deviation.

Parameters:

• sigma (float) –

  Standard deviation in element (voxel) space.

• truncate (float, default: 2.0 ) –

  The number of standard deviations to extend the kernel before truncating.

• device (device, default: None ) –

  The device on which to create the kernel.

• dtype (dtype, default: None ) –

  The kernel datatype.

Returns:

• Tensor –

  A normalized kernel of length \(2 * int(truncate * sigma + 0.5) + 1\).