kornia.enhance¶

The functions in this section perform normalisations and intensity transformations.

adjust_brightness(input: torch.Tensor, brightness_factor: Union[float, torch.Tensor]) → torch.Tensor[source]¶

Adjust Brightness of an image.

See AdjustBrightness for details.

adjust_contrast(input: torch.Tensor, contrast_factor: Union[float, torch.Tensor]) → torch.Tensor[source]¶

Adjust Contrast of an image.

See AdjustContrast for details.

adjust_gamma(input: torch.Tensor, gamma: Union[float, torch.Tensor], gain: Union[float, torch.Tensor] = 1.0) → torch.Tensor[source]¶

Perform gamma correction on an image.

See AdjustGamma for details.

adjust_hue(input: torch.Tensor, hue_factor: Union[float, torch.Tensor]) → torch.Tensor[source]¶

Adjust hue of an image.

See AdjustHue for details.

adjust_saturation(input: torch.Tensor, saturation_factor: Union[float, torch.Tensor]) → torch.Tensor[source]¶

Adjust color saturation of an image.

See AdjustSaturation for details.

add_weighted(src1: torch.Tensor, alpha: float, src2: torch.Tensor, beta: float, gamma: float) → torch.Tensor[source]¶

Blend two Tensors.

See AddWeighted for details.

normalize(data: torch.Tensor, mean: Union[torch.Tensor, float], std: Union[torch.Tensor, float]) → torch.Tensor[source]¶

Normalise the image with channel-wise mean and standard deviation.

See Normalize for details.

Parameters

data (torch.Tensor) – The image tensor to be normalised.
mean (torch.Tensor or float) – Mean for each channel.
std (torch.Tensor or float) – Standard deviations for each channel.

Returns

The normalised image tensor.

Return type

torch.Tensor

normalize_min_max(x: torch.Tensor, min_val: float = 0.0, max_val: float = 1.0, eps: float = 1e-06) → torch.Tensor[source]¶

Normalise an image tensor by MinMax and re-scales the value between a range.

The data is normalised using the following formulation:

\[y_i = (b - a) * \frac{x_i - \text{min}(x)}{\text{max}(x) - \text{min}(x)} + a\]

where \(a\) is \(\text{min_val}\) and \(b\) is \(\text{max_val}\).

Parameters

x (torch.Tensor) – The image tensor to be normalised with shape \((B, C, H, W)\).
min_val (float) – The minimum value for the new range. Default: 0.
max_val (float) – The maximum value for the new range. Default: 1.
eps (float) – Float number to avoid zero division. Default: 1e-6.

Returns

The normalised image tensor with same shape as input \((B, C, H, W)\).

Return type

torch.Tensor

denormalize(data: torch.Tensor, mean: Union[torch.Tensor, float], std: Union[torch.Tensor, float]) → torch.Tensor[source]¶

Denormalize the image given channel-wise mean and standard deviation.

See Normalize for details.

Parameters

data (torch.Tensor) – The image tensor to be normalised.
mean (torch.Tensor or float) – Mean for each channel.
std (torch.Tensor or float) – Standard deviations for each channel.

Returns

The normalised image tensor.

Return type

torch.Tensor

zca_mean(inp: torch.Tensor, dim: int = 0, unbiased: bool = True, eps: float = 1e-06, return_inverse: bool = False) → Tuple[torch.Tensor, torch.Tensor, Optional[torch.Tensor]][source]¶

Computes the ZCA whitening matrix and mean vector. The output can be used with linear_transform()

See ZCAWhitening for details.

Parameters

inp (torch.Tensor) – input data tensor
dim (int) – Specifies the dimension that serves as the samples dimension. Default = 0
unbiased (bool) – Whether to use the unbiased estimate of the covariance matrix. Default = True
eps (float) – a small number used for numerical stability. Default = 0
return_inverse (bool) – Whether to return the inverse ZCA transform.

shapes:

inp: \((D_0,...,D_{\text{dim}},...,D_N)\) is a batch of N-D tensors.
transform_matrix: \((\Pi_{d=0,d\neq \text{dim}}^N D_d, \Pi_{d=0,d\neq \text{dim}}^N D_d)\)
mean_vector: \((1, \Pi_{d=0,d\neq \text{dim}}^N D_d)\)
inv_transform: same shape as the transform matrix

Returns: A tuple containing the ZCA matrix and the mean vector. If return_inverse is set to True, then it returns the inverse ZCA matrix, otherwise it returns None.
Return type: Tuple[torch.Tensor, torch.Tensor, torch.Tensor]

Examples

>>> from kornia.color import zca_mean
>>> x = torch.tensor([[0,1],[1,0],[-1,0],[0,-1]], dtype = torch.float32)
>>> transform_matrix, mean_vector,_ = zca_mean(x) # Returns transformation matrix and data mean
>>> x = torch.rand(3,20,2,2)
>>> transform_matrix, mean_vector, inv_transform = zca_mean(x, dim = 1, return_inverse = True)
>>> # transform_matrix.size() equals (12,12) and the mean vector.size equal (1,12)

zca_whiten(inp: torch.Tensor, dim: int = 0, unbiased: bool = True, eps: float = 1e-06) → torch.Tensor[source]¶

Applies ZCA whitening transform.

See ZCAWhitening for details.

Parameters

inp (torch.Tensor) – input data tensor
dim (int) – Specifies the dimension that serves as the samples dimension. Default = 0
unbiased (bool) – Whether to use the unbiased estimate of the covariance matrix. Default = True
eps (float) – a small number used for numerial stablility. Default = 0

Returns

Whiten Input data

Return type

torch.Tensor

Examples

>>> import torch
>>> import kornia
>>> x = torch.tensor([[0,1],[1,0],[-1,0]], dtype = torch.float32)
>>> x_whiten = kornia.color.zca_whiten(x)

linear_transform(inp: torch.Tensor, transform_matrix: torch.Tensor, mean_vector: torch.Tensor, dim: int = 0) → torch.Tensor[source]¶

Given a transformation matrix and a mean vector, this function will flatten the input tensor along the given dimension and subtract the mean vector from it. Then the dot product with the transformation matrix will be computed and then the resulting tensor is reshaped to the original input shape.

\[\mathbf{X}_{T} = (\mathbf{X - \mu})(T)\]

Parameters

inp (torch.Tensor) – Input data \(X\)
transform_matrix (torch.Tensor) – Transform matrix \(T\)
mean_vector (torch.Tensor) – mean vector \(\mu\)
dim (int) – Batch dimension. Default = 0

shapes:

inp: \((D_0,...,D_{\text{dim}},...,D_N)\) is a batch of N-D tensors.
transform_matrix: \((\Pi_{d=0,d\neq \text{dim}}^N D_d, \Pi_{d=0,d\neq \text{dim}}^N D_d)\)
mean_vector: \((1, \Pi_{d=0,d\neq \text{dim}}^N D_d)\)

Returns: Transformed data
Return type: torch.Tensor

Example

>>> # Example where dim = 3
>>> inp = torch.ones((10,3,4,5))
>>> transform_mat = torch.ones((10*3*4,10*3*4))
>>> mean = 2*torch.ones((1,10*3*4))
>>> out = kornia.color.linear_transform(inp, transform_mat, mean, 3)
>>> print(out) # Should a be (10,3,4,5) tensor of -120s
>>> # Example where dim = 0
>>> inp = torch.ones((10,2))
>>> transform_mat = torch.ones((2,2))
>>> mean = torch.zeros((1,2))
>>> out = kornia.color.linear_transform(inp, transform_mat, mean)
>>> print(out) # Should a be (10,3,4,5) tensor of 2s

histogram(x: torch.Tensor, bins: torch.Tensor, bandwidth: torch.Tensor, epsilon: float = 1e-10) → torch.Tensor[source]¶

Function that estimates the histogram of the input tensor. The calculation uses kernel density estimation which requires a bandwidth (smoothing) parameter. :param x: (torch.Tensor), shape [BxN] :param bins: (torch.Tensor), shape [NUM_BINS] :param bandwidth: (torch.Tensor), shape [1], gaussian smoothing factor :param epsilon: (float), scalar, for numerical stability

Returns: (torch.Tensor), shape [BxNUM_BINS]
Return type: pdf

histogram2d(x1: torch.Tensor, x2: torch.Tensor, bins: torch.Tensor, bandwidth: torch.Tensor, epsilon: float = 1e-10) → torch.Tensor[source]¶

Function that estimates the histogram of the input tensor. The calculation uses kernel density estimation which requires a bandwidth (smoothing) parameter. :param x1: (torch.Tensor), shape [BxN1] :param x2: (torch.Tensor), shape [BxN2] :param bins: (torch.Tensor), shape [NUM_BINS] :param bandwidth: (torch.Tensor), scalar, gaussian smoothing factor :param epsilon: (float), scalar, for numerical stability

Returns: (torch.Tensor), shape [BxNUM_BINSxNUM_BINS]
Return type: pdf

solarize(input: torch.Tensor, thresholds: Union[float, torch.Tensor] = 0.5, additions: Union[float, torch.Tensor, None] = None) → torch.Tensor[source]¶

For each pixel in the image less than threshold, we add ‘addition’ amount to it and then clip the pixel value to be between 0 and 1.0. The value of ‘addition’ is between -0.5 and 0.5.

Parameters

input (torch.Tensor) – image tensor with shapes like (C, H, W) or (B, C, H, W) to solarize.
thresholds (float or torch.Tensor) – solarize thresholds. If int or one element tensor, input will be solarized across the whole batch. If 1-d tensor, input will be solarized element-wise, len(thresholds) == len(input).
additions (optional, float or torch.Tensor) – between -0.5 and 0.5. Default None. If None, no addition will be performed. If int or one element tensor, same addition will be added across the whole batch. If 1-d tensor, additions will be added element-wisely, len(additions) == len(input).

Returns

Solarized images.

Return type

torch.Tensor

posterize(input: torch.Tensor, bits: Union[int, torch.Tensor]) → torch.Tensor[source]¶

Reduce the number of bits for each color channel. Non-differentiable function, uint8 involved.

Parameters

input (torch.Tensor) – image tensor with shapes like (C, H, W) or (B, C, H, W) to posterize.
bits (int or torch.Tensor) – number of high bits. Must be in range [0, 8]. If int or one element tensor, input will be posterized by this bits. If 1-d tensor, input will be posterized element-wisely, len(bits) == input.shape[1]. If n-d tensor, input will be posterized element-channel-wisely, bits.shape == input.shape[:len(bits.shape)]

Returns

Image with reduced color channels.

Return type

torch.Tensor

sharpness(input: torch.Tensor, factor: Union[float, torch.Tensor]) → torch.Tensor[source]¶

Implements Sharpness function from PIL using torch ops.

Parameters

input (torch.Tensor) – image tensor with shapes like (C, H, W) or (B, C, H, W) to sharpen.
factor (float or torch.Tensor) – factor of sharpness strength. Must be above 0. If float or one element tensor, input will be sharpened by the same factor across the whole batch. If 1-d tensor, input will be sharpened element-wisely, len(factor) == len(input).

Returns

Sharpened image or images.

Return type

torch.Tensor

equalize(input: torch.Tensor) → torch.Tensor[source]¶

Apply equalize on the input tensor. Implements Equalize function from PIL using PyTorch ops based on uint8 format: https://github.com/tensorflow/tpu/blob/master/models/official/efficientnet/autoaugment.py#L352

Parameters: input (torch.Tensor) – image tensor with shapes like :math:(C, H, W) or :math:(B, C, H, W) to equalize.
Returns: Sharpened image or images.
Return type: torch.Tensor

Modules¶

class Normalize(mean: Union[torch.Tensor, float], std: Union[torch.Tensor, float])[source]¶

Normalize a tensor image or a batch of tensor images with mean and standard deviation.

Input must be a tensor of shape (C, H, W) or a batch of tensors \((*, C, H, W)\).

Given mean: (M1,...,Mn) and std: (S1,..,Sn) for n channels, this transform will normalize each channel of the input torch.Tensor i.e. input[channel] = (input[channel] - mean[channel]) / std[channel]

Parameters

mean (torch.Tensor or float) – Mean for each channel.
std (torch.Tensor or float) – Standard deviations for each channel.

class Denormalize(mean: Union[torch.Tensor, float], std: Union[torch.Tensor, float])[source]¶

Denormalize a tensor image or a batch of tensor images.

Input must be a tensor of shape (C, H, W) or a batch of tensors \((*, C, H, W)\).

Given mean: (M1,...,Mn) and std: (S1,..,Sn) for n channels, this transform will denormalize each channel of the input torch.Tensor i.e. input[channel] = (input[channel] * std[channel]) + mean[channel]

Parameters

mean (torch.Tensor or float) – Mean for each channel.
std (torch.Tensor or float) – Standard deviations for each channel.

class ZCAWhitening(dim: int = 0, eps: float = 1e-06, unbiased: bool = True, detach_transforms: bool = True, compute_inv: bool = False)[source]¶

Computes the ZCA whitening matrix transform and the mean vector and applies the transform to the data. The data tensor is flattened, and the mean \(\mathbf{\mu}\) and covariance matrix \(\mathbf{\Sigma}\) are computed from the flattened data \(\mathbf{X} \in \mathbb{R}^{N \times D}\), where \(N\) is the sample size and \(D\) is flattened dimensionality (e.g. for a tensor with size 5x3x2x2 \(N = 5\) and \(D = 12\)). The ZCA whitening transform is given by:

\[\mathbf{X}_{\text{zca}} = (\mathbf{X - \mu})(US^{-\frac{1}{2}}U^T)^T\]

where \(U\) are the eigenvectors of \(\Sigma\) and \(S\) contain the correpsonding eigenvalues of \(\Sigma\). After the transform is applied, the output is reshaped to same shape.

Parameters

dim (int) – Determines the dimension that represents the samples axis. Default = 0
eps (float) – a small number used for numerial stablility. Default=1e-6
unbiased (bool) – Whether to use the biased estimate of the covariance matrix. Default=False
compute_inv (bool) – Compute the inverse transform matrix. Default=False
detach_transforms (bool) – Detaches gradient from the ZCA fitting. Default=True

shape:

x: \((D_0,...,D_{\text{dim}},...,D_N)\) is a batch of N-D tensors.
x_whiten: \((D_0,...,D_{\text{dim}},...,D_N)\) same shape as input.

Examples

>>> x = torch.tensor([[0,1],[1,0],[-1,0],[0,-1]], dtype = torch.float32)
>>> zca = kornia.color.ZCAWhitening().fit(x)
>>> x_whiten = zca(x)
>>> zca = kornia.color.ZCAWhitening()
>>> x_whiten = zca(x, include_fit = True) # Includes the fitting step
>>> x_whiten = zca(x) # Can run now without the fitting set
>>> # Enable backprop through ZCA fitting process
>>> zca = kornia.color.ZCAWhitening(detach_transforms = False)
>>> x_whiten = zca(x, include_fit = True) # Includes the fitting step

Note

This implementation uses svd() which yields NaNs in the backwards step if the sigular values are not unique. See here for more information.

References

[1] Stanford PCA & ZCA whitening tutorial

fit(x: torch.Tensor)[source]¶

Fits ZCA whitening matrices to the data.

Parameters: x (torch.Tensor) – Input data
Returns: returns a fitted ZCAWhiten object instance.
Return type: ZCAWhiten

forward(x: torch.Tensor, include_fit: bool = False) → torch.Tensor[source]¶

Applies the whitening transform to the data

Parameters

x (torch.Tensor) – Input data
include_fit (bool) – Indicates whether to fit the data as part of the forward pass

Returns

The transformed data

Return type

torch.Tensor

inverse_transform(x: torch.Tensor) → torch.Tensor[source]¶

Applies the inverse transform to the whitened data.

Parameters: x (torch.Tensor) – Whitened data
Returns: original data
Return type: torch.Tensor

class AdjustBrightness(brightness_factor: Union[float, torch.Tensor])[source]¶

Adjust Brightness of an image. This implementation aligns OpenCV, not PIL. Hence, the output differs from TorchVision.

The input image is expected to be in the range of [0, 1].

Parameters

input (torch.Tensor) – Image/Input to be adjusted in the shape of (*, N).
brightness_factor (Union[float, torch.Tensor]) – Brightness adjust factor per element in the batch. 0 does not modify the input image while any other number modify the brightness.

Returns

Adjusted image.

Return type

torch.Tensor

class AdjustContrast(contrast_factor: Union[float, torch.Tensor])[source]¶

Adjust Contrast of an image. This implementation aligns OpenCV, not PIL. Hence, the output differs from TorchVision.

The input image is expected to be in the range of [0, 1].

Parameters

input (torch.Tensor) – Image to be adjusted in the shape of (*, N).
contrast_factor (Union[float, torch.Tensor]) – Contrast adjust factor per element in the batch. 0 generates a compleatly black image, 1 does not modify the input image while any other non-negative number modify the brightness by this factor.

Returns

Adjusted image.

Return type

torch.Tensor

class AdjustSaturation(saturation_factor: Union[float, torch.Tensor])[source]¶

Adjust color saturation of an image.

The input image is expected to be an RGB image in the range of [0, 1].

Parameters

input (torch.Tensor) – Image/Tensor to be adjusted in the shape of (*, N).
saturation_factor (float) – How much to adjust the saturation. 0 will give a black
white image, 1 will give the original image while 2 will enhance the saturation (and) –
a factor of 2. (by) –

Returns

Adjusted image.

Return type

torch.Tensor

class AdjustHue(hue_factor: Union[float, torch.Tensor])[source]¶

Adjust hue of an image.

The input image is expected to be an RGB image in the range of [0, 1].

Parameters

input (torch.Tensor) – Image/Tensor to be adjusted in the shape of (*, N).
hue_factor (float) – How much to shift the hue channel. Should be in [-PI, PI]. PI and -PI give complete reversal of hue channel in HSV space in positive and negative direction respectively. 0 means no shift. Therefore, both -PI and PI will give an image with complementary colors while 0 gives the original image.

Returns

Adjusted image.

Return type

torch.Tensor

class AdjustGamma(gamma: Union[float, torch.Tensor], gain: Union[float, torch.Tensor] = 1.0)[source]¶

Perform gamma correction on an image.

The input image is expected to be in the range of [0, 1].

Parameters

input (torch.Tensor) – Image/Tensor to be adjusted in the shape of (*, N).
gamma (float) – Non negative real number, same as γgammaγ in the equation. gamma larger than 1 make the shadows darker, while gamma smaller than 1 make dark regions lighter.
gain (float, optional) – The constant multiplier. Default 1.

Returns

Adjusted image.

Return type

torch.Tensor

class AddWeighted(alpha: float, beta: float, gamma: float)[source]¶

Calculates the weighted sum of two Tensors.

The function calculates the weighted sum of two Tensors as follows:

\[out = src1 * alpha + src2 * beta + gamma\]

Parameters

src1 (torch.Tensor) – Tensor.
alpha (float) – weight of the src1 elements.
src2 (torch.Tensor) – Tensor of same size and channel number as src1.
beta (float) – weight of the src2 elements.
gamma (float) – scalar added to each sum.

Returns

Weighted Tensor.

Return type

torch.Tensor