kornia.geometry.conversions

rad2deg(tensor)[source]

Function that converts angles from radians to degrees.

Parameters

tensor (Tensor) – Tensor of arbitrary shape.

Return type

Tensor

Returns

Tensor with same shape as input.

Example

>>> input = torch.tensor(3.1415926535) * torch.rand(1, 3, 3)
>>> output = rad2deg(input)
deg2rad(tensor)[source]

Function that converts angles from degrees to radians.

Parameters

tensor (Tensor) – Tensor of arbitrary shape.

Return type

Tensor

Returns

tensor with same shape as input.

Examples

>>> input = 360. * torch.rand(1, 3, 3)
>>> output = deg2rad(input)
pol2cart(rho, phi)[source]

Function that converts polar coordinates to cartesian coordinates.

Parameters
  • rho (Tensor) – Tensor of arbitrary shape.

  • phi (Tensor) – Tensor of same arbitrary shape.

Return type

Tuple[Tensor, Tensor]

Returns

Tensor with same shape as input.

Example

>>> rho = torch.rand(1, 3, 3)
>>> phi = torch.rand(1, 3, 3)
>>> x, y = pol2cart(rho, phi)
cart2pol(x, y, eps=1e-08)[source]

Function that converts cartesian coordinates to polar coordinates.

Parameters
  • rho – Tensor of arbitrary shape.

  • phi – Tensor of same arbitrary shape.

  • eps (float, optional) – To avoid division by zero. Default: 1e-08

Return type

Tuple[Tensor, Tensor]

Returns

Tensor with same shape as input.

Example

>>> x = torch.rand(1, 3, 3)
>>> y = torch.rand(1, 3, 3)
>>> rho, phi = cart2pol(x, y)
convert_points_from_homogeneous(points, eps=1e-08)[source]

Function that converts points from homogeneous to Euclidean space.

Parameters
  • points (Tensor) – the points to be transformed.

  • eps (float, optional) – to avoid division by zero. Default: 1e-08

Return type

Tensor

Returns

the points in Euclidean space.

Examples

>>> input = torch.rand(2, 4, 3)  # BxNx3
>>> output = convert_points_from_homogeneous(input)  # BxNx2
convert_points_to_homogeneous(points)[source]

Function that converts points from Euclidean to homogeneous space.

Parameters

points (Tensor) – the points to be transformed.

Return type

Tensor

Returns

the points in homogeneous coordinates.

Examples

>>> input = torch.rand(2, 4, 3)  # BxNx3
>>> output = convert_points_to_homogeneous(input)  # BxNx4
convert_affinematrix_to_homography(A)[source]

Function that converts batch of affine matrices.

Parameters

A (Tensor) – the affine matrix with shape \((B,2,3)\).

Return type

Tensor

Returns

the homography matrix with shape of \((B,3,3)\).

Examples

>>> input = torch.rand(2, 2, 3)  # Bx2x3
>>> output = convert_affinematrix_to_homography(input)  # Bx3x3
rotation_matrix_to_angle_axis(rotation_matrix)[source]

Convert 3x3 rotation matrix to Rodrigues vector.

Parameters

rotation_matrix (Tensor) – rotation matrix.

Return type

Tensor

Returns

Rodrigues vector transformation.

Shape:
  • Input: \((N, 3, 3)\)

  • Output: \((N, 3)\)

Example

>>> input = torch.rand(2, 3, 3)  # Nx3x3
>>> output = rotation_matrix_to_angle_axis(input)  # Nx3
rotation_matrix_to_quaternion(rotation_matrix, eps=1e-08, order=QuaternionCoeffOrder.XYZW)[source]

Convert 3x3 rotation matrix to 4d quaternion vector.

The quaternion vector has components in (w, x, y, z) or (x, y, z, w) format.

Note

The (x, y, z, w) order is going to be deprecated in favor of efficiency.

Parameters
  • rotation_matrix (Tensor) – the rotation matrix to convert.

  • eps (float, optional) – small value to avoid zero division. Default: 1e-08

  • order (QuaternionCoeffOrder, optional) – quaternion coefficient order. Note: ‘xyzw’ will be deprecated in favor of ‘wxyz’. Default: QuaternionCoeffOrder.XYZW

Return type

Tensor

Returns

the rotation in quaternion.

Shape:
  • Input: \((*, 3, 3)\)

  • Output: \((*, 4)\)

Example

>>> input = torch.rand(4, 3, 3)  # Nx3x3
>>> output = rotation_matrix_to_quaternion(input, eps=torch.finfo(input.dtype).eps,
...                                        order=QuaternionCoeffOrder.WXYZ)  # Nx4
quaternion_to_angle_axis(quaternion, order=QuaternionCoeffOrder.XYZW)[source]

Convert quaternion vector to angle axis of rotation.

The quaternion should be in (x, y, z, w) or (w, x, y, z) format.

Adapted from ceres C++ library: ceres-solver/include/ceres/rotation.h

Parameters
  • quaternion (Tensor) – tensor with quaternions.

  • order (QuaternionCoeffOrder, optional) – quaternion coefficient order. Note: ‘xyzw’ will be deprecated in favor of ‘wxyz’. Default: QuaternionCoeffOrder.XYZW

Return type

Tensor

Returns

tensor with angle axis of rotation.

Shape:
  • Input: \((*, 4)\) where * means, any number of dimensions

  • Output: \((*, 3)\)

Example

>>> quaternion = torch.rand(2, 4)  # Nx4
>>> angle_axis = quaternion_to_angle_axis(quaternion)  # Nx3
quaternion_to_rotation_matrix(quaternion, order=QuaternionCoeffOrder.XYZW)[source]

Converts a quaternion to a rotation matrix.

The quaternion should be in (x, y, z, w) or (w, x, y, z) format.

Parameters
  • quaternion (Tensor) – a tensor containing a quaternion to be converted. The tensor can be of shape \((*, 4)\).

  • order (QuaternionCoeffOrder, optional) – quaternion coefficient order. Note: ‘xyzw’ will be deprecated in favor of ‘wxyz’. Default: QuaternionCoeffOrder.XYZW

Return type

Tensor

Returns

the rotation matrix of shape \((*, 3, 3)\).

Example

>>> quaternion = torch.tensor((0., 0., 0., 1.))
>>> quaternion_to_rotation_matrix(quaternion, order=QuaternionCoeffOrder.WXYZ)
tensor([[-1.,  0.,  0.],
        [ 0., -1.,  0.],
        [ 0.,  0.,  1.]])
quaternion_log_to_exp(quaternion, eps=1e-08, order=QuaternionCoeffOrder.XYZW)[source]

Applies exponential map to log quaternion.

The quaternion should be in (x, y, z, w) or (w, x, y, z) format.

Parameters
  • quaternion (Tensor) – a tensor containing a quaternion to be converted. The tensor can be of shape \((*, 3)\).

  • order (QuaternionCoeffOrder, optional) – quaternion coefficient order. Note: ‘xyzw’ will be deprecated in favor of ‘wxyz’. Default: QuaternionCoeffOrder.XYZW

Return type

Tensor

Returns

the quaternion exponential map of shape \((*, 4)\).

Example

>>> quaternion = torch.tensor((0., 0., 0.))
>>> quaternion_log_to_exp(quaternion, eps=torch.finfo(quaternion.dtype).eps,
...                       order=QuaternionCoeffOrder.WXYZ)
tensor([1., 0., 0., 0.])
quaternion_exp_to_log(quaternion, eps=1e-08, order=QuaternionCoeffOrder.XYZW)[source]

Applies the log map to a quaternion.

The quaternion should be in (x, y, z, w) format.

Parameters
  • quaternion (Tensor) – a tensor containing a quaternion to be converted. The tensor can be of shape \((*, 4)\).

  • eps (float, optional) – A small number for clamping. Default: 1e-08

  • order (QuaternionCoeffOrder, optional) – quaternion coefficient order. Note: ‘xyzw’ will be deprecated in favor of ‘wxyz’. Default: QuaternionCoeffOrder.XYZW

Return type

Tensor

Returns

the quaternion log map of shape \((*, 3)\).

Example

>>> quaternion = torch.tensor((1., 0., 0., 0.))
>>> quaternion_exp_to_log(quaternion, eps=torch.finfo(quaternion.dtype).eps,
...                       order=QuaternionCoeffOrder.WXYZ)
tensor([0., 0., 0.])
angle_axis_to_quaternion(angle_axis, order=QuaternionCoeffOrder.XYZW)[source]

Convert an angle axis to a quaternion.

The quaternion vector has components in (x, y, z, w) or (w, x, y, z) format.

Adapted from ceres C++ library: ceres-solver/include/ceres/rotation.h

Parameters
  • angle_axis (Tensor) – tensor with angle axis.

  • order (QuaternionCoeffOrder, optional) – quaternion coefficient order. Note: ‘xyzw’ will be deprecated in favor of ‘wxyz’. Default: QuaternionCoeffOrder.XYZW

Return type

Tensor

Returns

tensor with quaternion.

Shape:
  • Input: \((*, 3)\) where * means, any number of dimensions

  • Output: \((*, 4)\)

Example

>>> angle_axis = torch.rand(2, 3)  # Nx3
>>> quaternion = angle_axis_to_quaternion(angle_axis, order=QuaternionCoeffOrder.WXYZ)  # Nx4
angle_axis_to_rotation_matrix(angle_axis)[source]

Convert 3d vector of axis-angle rotation to 3x3 rotation matrix.

Parameters

angle_axis (Tensor) – tensor of 3d vector of axis-angle rotations.

Return type

Tensor

Returns

tensor of 3x3 rotation matrices.

Shape:
  • Input: \((N, 3)\)

  • Output: \((N, 3, 3)\)

Example

>>> input = torch.rand(1, 3)  # Nx3
>>> output = angle_axis_to_rotation_matrix(input)  # Nx3x3
denormalize_pixel_coordinates(pixel_coordinates, height, width, eps=1e-08)[source]

Denormalize pixel coordinates.

The input is assumed to be -1 if on extreme left, 1 if on extreme right (x = w-1).

Parameters
  • pixel_coordinates (Tensor) – the normalized grid coordinates. Shape can be \((*, 2)\).

  • width (int) – the maximum width in the x-axis.

  • height (int) – the maximum height in the y-axis.

  • eps (float, optional) – safe division by zero. Default: 1e-08

Return type

Tensor

Returns

the denormalized pixel coordinates.

normalize_pixel_coordinates(pixel_coordinates, height, width, eps=1e-08)[source]

Normalize pixel coordinates between -1 and 1.

Normalized, -1 if on extreme left, 1 if on extreme right (x = w-1).

Parameters
  • pixel_coordinates (Tensor) – the grid with pixel coordinates. Shape can be \((*, 2)\).

  • width (int) – the maximum width in the x-axis.

  • height (int) – the maximum height in the y-axis.

  • eps (float, optional) – safe division by zero. Default: 1e-08

Return type

Tensor

Returns

the normalized pixel coordinates.

denormalize_pixel_coordinates3d(pixel_coordinates, depth, height, width, eps=1e-08)[source]

Denormalize pixel coordinates.

The input is assumed to be -1 if on extreme left, 1 if on extreme right (x = w-1).

Parameters
  • pixel_coordinates (Tensor) – the normalized grid coordinates. Shape can be \((*, 3)\).

  • depth (int) – the maximum depth in the x-axis.

  • height (int) – the maximum height in the y-axis.

  • width (int) – the maximum width in the x-axis.

  • eps (float, optional) – safe division by zero. Default: 1e-08

Return type

Tensor

Returns

the denormalized pixel coordinates.

normalize_pixel_coordinates3d(pixel_coordinates, depth, height, width, eps=1e-08)[source]

Normalize pixel coordinates between -1 and 1.

Normalized, -1 if on extreme left, 1 if on extreme right (x = w-1).

Parameters
  • pixel_coordinates (Tensor) – the grid with pixel coordinates. Shape can be \((*, 3)\).

  • depth (int) – the maximum depth in the z-axis.

  • height (int) – the maximum height in the y-axis.

  • width (int) – the maximum width in the x-axis.

  • eps (float, optional) – safe division by zero. Default: 1e-08

Return type

Tensor

Returns

the normalized pixel coordinates.

normalize_quaternion(quaternion, eps=1e-12)[source]

Normalizes a quaternion.

The quaternion should be in (x, y, z, w) format.

Parameters
  • quaternion (Tensor) – a tensor containing a quaternion to be normalized. The tensor can be of shape \((*, 4)\).

  • eps (float, optional) – small value to avoid division by zero. Default: 1e-12

Return type

Tensor

Returns

the normalized quaternion of shape \((*, 4)\).

Example

>>> quaternion = torch.tensor((1., 0., 1., 0.))
>>> normalize_quaternion(quaternion)
tensor([0.7071, 0.0000, 0.7071, 0.0000])