kornia.geometry.transform¶

The functions in this section perform various geometrical transformations of 2D images.

warp_perspective(src, M, dsize, flags='bilinear', border_mode=None, border_value=0)[source]¶

Applies a perspective transformation to an image.

The function warp_perspective transforms the source image using the specified matrix:

\[\text{dst} (x, y) = \text{src} \left( \frac{M_{11} x + M_{12} y + M_{13}}{M_{31} x + M_{32} y + M_{33}} , \frac{M_{21} x + M_{22} y + M_{23}}{M_{31} x + M_{32} y + M_{33}} \right )\]

Parameters:	src (torch.Tensor) – input image. M (Tensor) – transformation matrix. dsize (tuple) – size of the output image (height, width).
Returns:	the warped input image.
Return type:	Tensor

Shape:

Input: \((B, C, H, W)\) and \((B, 3, 3)\)
Output: \((B, C, H, W)\)

Note

See a working example here.

warp_affine(src: torch.Tensor, M: torch.Tensor, dsize: Tuple[int, int], flags: Optional[str] = 'bilinear', padding_mode: Optional[str] = 'zeros') → torch.Tensor[source]¶

Applies an affine transformation to a tensor.

The function warp_affine transforms the source tensor using the specified matrix:

\[\text{dst}(x, y) = \text{src} \left( M_{11} x + M_{12} y + M_{13} , M_{21} x + M_{22} y + M_{23} \right )\]

Parameters:	src (torch.Tensor) – input tensor of shape \((B, C, H, W)\). M (torch.Tensor) – affine transformation of shape \((B, 2, 3)\). dsize (Tuple[int, int]) – size of the output image (height, width). mode (Optional[str]) – interpolation mode to calculate output values ‘bilinear’ \| ‘nearest’. Default: ‘bilinear’. padding_mode (Optional[str]) – padding mode for outside grid values ‘zeros’ \| ‘border’ \| ‘reflection’. Default: ‘zeros’.
Returns:	the warped tensor.
Return type:	torch.Tensor

Shape:

Output: \((B, C, H, W)\)

Note

See a working example here.

get_perspective_transform(src, dst)[source]¶

Calculates a perspective transform from four pairs of the corresponding points.

The function calculates the matrix of a perspective transform so that:

\[\begin{split}\begin{bmatrix} t_{i}x_{i}^{'} \\ t_{i}y_{i}^{'} \\ t_{i} \\ \end{bmatrix} = \textbf{map_matrix} \cdot \begin{bmatrix} x_{i} \\ y_{i} \\ 1 \\ \end{bmatrix}\end{split}\]

where

\[dst(i) = (x_{i}^{'},y_{i}^{'}), src(i) = (x_{i}, y_{i}), i = 0,1,2,3\]

Parameters:	src (Tensor) – coordinates of quadrangle vertices in the source image. dst (Tensor) – coordinates of the corresponding quadrangle vertices in the destination image.
Returns:	the perspective transformation.
Return type:	Tensor

Shape:

Input: \((B, 4, 2)\) and \((B, 4, 2)\)
Output: \((B, 3, 3)\)

get_rotation_matrix2d(center: torch.Tensor, angle: torch.Tensor, scale: torch.Tensor) → torch.Tensor[source]¶

Calculates an affine matrix of 2D rotation.

The function calculates the following matrix:

\[\begin{split}\begin{bmatrix} \alpha & \beta & (1 - \alpha) \cdot \text{x} - \beta \cdot \text{y} \\ -\beta & \alpha & \beta \cdot \text{x} + (1 - \alpha) \cdot \text{y} \end{bmatrix}\end{split}\]

where

\[\begin{split}\alpha = \text{scale} \cdot cos(\text{angle}) \\ \beta = \text{scale} \cdot sin(\text{angle})\end{split}\]

The transformation maps the rotation center to itself If this is not the target, adjust the shift.

Parameters:	center (Tensor) – center of the rotation in the source image. angle (Tensor) – rotation angle in degrees. Positive values mean counter-clockwise rotation (the coordinate origin is assumed to be the top-left corner). scale (Tensor) – isotropic scale factor.
Returns:	the affine matrix of 2D rotation.
Return type:	Tensor

Shape:

Input: \((B, 2)\), \((B)\) and \((B)\)
Output: \((B, 2, 3)\)

Example

>>> center = torch.zeros(1, 2)
>>> scale = torch.ones(1)
>>> angle = 45. * torch.ones(1)
>>> M = kornia.get_rotation_matrix2d(center, angle, scale)
tensor([[[ 0.7071,  0.7071,  0.0000],
         [-0.7071,  0.7071,  0.0000]]])

remap(tensor: torch.Tensor, map_x: torch.Tensor, map_y: torch.Tensor) → torch.Tensor[source]¶

Applies a generic geometrical transformation to a tensor.

The function remap transforms the source tensor using the specified map:

\[\text{dst}(x, y) = \text{src}(map_x(x, y), map_y(x, y))\]

Parameters:	tensor (torch.Tensor) – the tensor to remap with shape (B, D, H, W). Where D is the number of channels. map_x (torch.Tensor) – the flow in the x-direction in pixel coordinates. The tensor must be in the shape of (B, H, W). map_y (torch.Tensor) – the flow in the y-direction in pixel coordinates. The tensor must be in the shape of (B, H, W).
Returns:	the warped tensor.
Return type:	torch.Tensor

Example

>>> grid = kornia.utils.create_meshgrid(2, 2, False)  # 1x2x2x2
>>> grid += 1  # apply offset in both directions
>>> input = torch.ones(1, 1, 2, 2)
>>> kornia.remap(input, grid[..., 0], grid[..., 1])   # 1x1x2x2
tensor([[[[1., 0.],
          [0., 0.]]]])

invert_affine_transform(matrix: torch.Tensor) → torch.Tensor[source]¶

Inverts an affine transformation.

The function computes an inverse affine transformation represented by 2×3 matrix:

\[\begin{split}\begin{bmatrix} a_{11} & a_{12} & b_{1} \\ a_{21} & a_{22} & b_{2} \\ \end{bmatrix}\end{split}\]

The result is also a 2×3 matrix of the same type as M.

Parameters:	matrix (torch.Tensor) – original affine transform. The tensor musth be in the shape of (B, 2, 3).
Returns:	the reverse affine transform.
Return type:	torch.Tensor

center_crop(tensor: torch.Tensor, size: Tuple[int, int]) → torch.Tensor[source]¶

Crops the given tensor at the center.

Parameters:	tensor (torch.Tensor) – the input tensor with shape (C, H, W) or (B, C, H, W). size (Tuple[int, int]) – a tuple with the expected height and width of the output patch.
Returns:	the output tensor with patches.
Return type:	torch.Tensor

Examples

>>> input = torch.tensor([[
        [1., 2., 3., 4.],
        [5., 6., 7., 8.],
        [9., 10., 11., 12.],
        [13., 14., 15., 16.],
     ]])
>>> kornia.center_crop(input, (2, 4))
tensor([[[ 5.0000,  6.0000,  7.0000,  8.0000],
         [ 9.0000, 10.0000, 11.0000, 12.0000]]])

crop_and_resize(tensor: torch.Tensor, boxes: torch.Tensor, size: Tuple[int, int]) → torch.Tensor[source]¶

Extracts crops from the input tensor and resizes them.

Parameters:	tensor (torch.Tensor) – the reference tensor of shape BxCxHxW. boxes (torch.Tensor) – a tensor containing the coordinates of the bounding boxes to be extracted. The tensor must have the shape of Bx4x2, where each box is defined in the following order: top-left, top-right, bottom-left and bottom-right. The coordinates order must be in y, x respectively. size (Tuple[int, int]) – a tuple with the height and width that will be used to resize the extracted patches.
Returns:	tensor containing the patches with shape BxN1xN2
Return type:	torch.Tensor

Example

>>> input = torch.tensor([[
        [1., 2., 3., 4.],
        [5., 6., 7., 8.],
        [9., 10., 11., 12.],
        [13., 14., 15., 16.],
    ]])
>>> boxes = torch.tensor([[
        [1., 1.],
        [1., 2.],
        [2., 1.],
        [2., 2.],
    ]])  # 1x4x2
>>> kornia.crop_and_resize(input, boxes, (2, 2))
tensor([[[ 6.0000,  7.0000],
         [ 10.0000, 11.0000]]])

pyrdown(input: torch.Tensor, border_type: str = 'reflect') → torch.Tensor[source]¶

Blurs a tensor and downsamples it.

See PyrDown for details.

pyrup(input: torch.Tensor, border_type: str = 'reflect') → torch.Tensor[source]¶

Upsamples a tensor and then blurs it.

See PyrUp for details.

affine(tensor: torch.Tensor, matrix: torch.Tensor) → torch.Tensor[source]¶

Apply an affine transformation to the image.

Parameters:	tensor (torch.Tensor) – The image tensor to be warped. matrix (torch.Tensor) – The 2x3 affine transformation matrix.
Returns:	The warped image.
Return type:	torch.Tensor

rotate(tensor: torch.Tensor, angle: torch.Tensor, center: Union[None, torch.Tensor] = None) → torch.Tensor[source]¶

Rotate the image anti-clockwise about the centre.

See Rotate for details.

translate(tensor: torch.Tensor, translation: torch.Tensor) → torch.Tensor[source]¶

Translate the tensor in pixel units.

See Translate for details.

scale(tensor: torch.Tensor, scale_factor: torch.Tensor, center: Union[None, torch.Tensor] = None) → torch.Tensor[source]¶

Scales the input image.

See Scale for details.

shear(tensor: torch.Tensor, shear: torch.Tensor) → torch.Tensor[source]¶

Shear the tensor.

See Shear for details.

class Rotate(angle: torch.Tensor, center: Union[None, torch.Tensor] = None)[source]¶

Rotate the tensor anti-clockwise about the centre.

Parameters:	angle (torch.Tensor) – The angle through which to rotate. The tensor must have a shape of (B), where B is batch size. center (torch.Tensor) – The center through which to rotate. The tensor must have a shape of (B, 2), where B is batch size and last dimension contains cx and cy.
Returns:	The rotated tensor.
Return type:	torch.Tensor

class Translate(translation: torch.Tensor)[source]¶

Translate the tensor in pixel units.

Parameters:	translation (torch.Tensor) – tensor containing the amount of pixels to translate in the x and y direction. The tensor must have a shape of (B, 2), where B is batch size, last dimension contains dx dy.
Returns:	The translated tensor.
Return type:	torch.Tensor

class Scale(scale_factor: torch.Tensor, center: Union[None, torch.Tensor] = None)[source]¶

Scale the tensor by a factor.

Parameters:	scale_factor (torch.Tensor) – The scale factor apply. The tensor must have a shape of (B), where B is batch size. center (torch.Tensor) – The center through which to scale. The tensor must have a shape of (B, 2), where B is batch size and last dimension contains cx and cy.
Returns:	The scaled tensor.
Return type:	torch.Tensor

class Shear(shear: torch.Tensor)[source]¶

Shear the tensor.

Parameters:	tensor (torch.Tensor) – The image tensor to be skewed. shear (torch.Tensor) – tensor containing the angle to shear in the x and y direction. The tensor must have a shape of (B, 2), where B is batch size, last dimension contains shx shy.
Returns:	The skewed tensor.
Return type:	torch.Tensor

class PyrDown(border_type: str = 'reflect')[source]¶

Blurs a tensor and downsamples it.

Parameters:	borde_type (str) – the padding mode to be applied before convolving. The expected modes are: `'constant'`, `'reflect'`, `'replicate'` or `'circular'`. Default: `'reflect'`.
Returns:	the downsampled tensor.
Return type:	torch.Tensor

Shape:

Input: \((B, C, H, W)\)
Output: \((B, C, H / 2, W / 2)\)

Examples

>>> input = torch.rand(1, 2, 4, 4)
>>> output = kornia.transform.PyrDown()(input)  # 1x2x2x2

class PyrUp(border_type: str = 'reflect')[source]¶

Upsamples a tensor and then blurs it.

Parameters:	borde_type (str) – the padding mode to be applied before convolving. The expected modes are: `'constant'`, `'reflect'`, `'replicate'` or `'circular'`. Default: `'reflect'`.
Returns:	the upsampled tensor.
Return type:	torch.Tensor

Shape:

Input: \((B, C, H, W)\)
Output: \((B, C, H * 2, W * 2)\)

Examples

>>> input = torch.rand(1, 2, 4, 4)
>>> output = kornia.transform.PyrUp()(input)  # 1x2x8x8

class ScalePyramid(n_levels: int = 3, init_sigma: float = 1.6, min_size: int = 5)[source]¶

Creates an scale pyramid of image, usually used for local feature detection. Images are consequently smoothed with Gaussian blur and downscaled. :param n_levels: number of the levels in octave. :type n_levels: int :param init_sigma: initial blur level. :type init_sigma: float :param min_size: the minimum size of the octave in pixels. Default is 5 :type min_size: int

Returns:	1st output: images 2nd output: sigmas (coefficients for scale conversion) 3rd output: pixelDists (coefficients for coordinate conversion)
Return type:	Tuple(List(Tensors), List(Tensors), List(Tensors))

Shape:

Input: \((B, C, H, W)\)
Output 1st: \([(B, NL, C, H, W), (B, NL, C, H/2, W/2), ...]\)
Output 2nd: \([(B, NL), (B, NL), (B, NL), ...]\)
Output 3rd: \([(B, NL), (B, NL), (B, NL), ...]\)

Examples::

>>> input = torch.rand(2, 4, 100, 100)
>>> sp, sigmas, pds = kornia.ScalePyramid(3, 15)(input)