Kornia

last updated on 10/10/2023

Repository Documentation Transformations Paper

Supported languages:

Python

Supported ML libraries:

PyTorch

Kornia is a differentiable computer vision library for PyTorch, with modules dedicated to image augmentation.

Transformations Description Tags

from kornia.enhance

Calculates the weighted sum of two image tensors.

MIXED SAMPLE

Original	Augmented


`out = K.enhance.AddWeighted(0.5, 0.5, 0)(img, img_2)`

from kornia.enhance

Adjusts log correction on the input image. Reference,

PHOTOMETRY

Original	Augmented

`aug = K.enhance.AdjustLog(gain=0.5, inv=False, clip_output=True)`

Original	Augmented

`aug = K.enhance.AdjustLog(gain=1.3, inv=False, clip_output=True)`

Original	Augmented

`aug = K.enhance.AdjustLog(gain=0.5, inv=True, clip_output=True)`

Original	Augmented

`aug = K.enhance.AdjustLog(gain=1.3, inv=True, clip_output=True)`

from kornia.enhance

Adjusts the contrast of the image or performs sigmoid correction on the input image. Reference.

PHOTOMETRY

Original	Augmented

`aug = K.enhance.AdjustSigmoid(cutoff=0.35, gain=7, inv=False)`

Original	Augmented

`aug = K.enhance.AdjustSigmoid(cutoff=0.45, gain=7, inv=False)`

from kornia.augmentation.auto

Applies AutoAugment searched strategies.

Original	Augmented

Original	Augmented

`aug = K.augmentation.auto.AutoAugment(policy='imagenet')`

from kornia.filters

Blurs an image using a Bilateral filter.

BLUR

Original	Augmented

`aug = K.filters.BilateralBlur(kernel_size=(5, 5), sigma_color=2.0, sigma_space=(1.5, 1.5))`

from kornia.filters

Computes blur (anti-aliasing) and downsamples a given feature map. Reference.

BLUR

Original	Augmented

`aug = K.filters.BlurPool2D(kernel_size=5, stride=3)`

from kornia.augmentation

Crops a given image at the center.

BASIC GEOMETRY

Original	Augmented

`aug = K.augmentation.CenterCrop(size=(1000,1000), p=1.0, cropping_mode='slice')`

from kornia.augmentation

Applies a random transformation to the brightness, contrast, saturation and hue of an image.

PHOTOMETRY

Original	Augmented

Original	Augmented

`aug = K.augmentation.ColorJiggle(brightness=0.2, contrast=0.1, saturation=0.4, hue=0.1, p=1.0)`

from kornia.augmentation

Applies a random transformation to the brightness, contrast, saturation and hue of an image.

PHOTOMETRY

Original	Augmented

Original	Augmented

`aug = K.augmentation.ColorJitter(brightness=0.2, contrast=0.1, saturation=0.4, hue=0.1, p=1.0)`

10.

from kornia.filters

Blurs an image using a Guided filter.

BLUR

11.

from kornia.filters

Blurs an image using a Joint Bilateral filter.

BLUR

12.

from kornia.augmentation

Rescales an image so that maximum side is equal to max_size, keeping the aspect ratio of the initial image.

BASIC GEOMETRY

Original	Augmented

`aug = K.augmentation.LongestMaxSize(max_size=400, p=1.0)`

13.

from kornia.filters

Computes pools and blurs and downsamples a given feature map.

BLUR

Original	Augmented

`aug = K.filters.MaxBlurPool2D(kernel_size=(5, 5))`

14.

from kornia.augmentation

Pads the given sample to a specific size.

BASIC GEOMETRY

Original	Augmented

`aug = K.augmentation.PadTo(size=(2000, 1500), pad_mode='constant', pad_value=0.5)`

Original	Augmented

`aug = K.augmentation.PadTo(size=(1500, 2000), pad_mode='constant', pad_value=0.5)`

15.

from kornia.augmentation.auto

Applies RandAugment augmentation strategies.

Original	Augmented

Original	Augmented

`aug = K.augmentation.auto.RandAugment(n=3, m=15)`

16.

from kornia.augmentation

Applies a random 2D affine transformation to an image.

BASIC GEOMETRY

Original	Augmented

`aug = K.augmentation.RandomAffine(degrees=(20, 20), translate=(0.2, 0.3), scale=None, shear=None, padding_mode='reflection', p=1)`

17.

from kornia.augmentation

Adds random blur with a box filter to an image.

BLUR

Original	Augmented

`aug = K.augmentation.RandomBoxBlur(kernel_size=(15, 15), p=1)`

18.

from kornia.augmentation

Applies a random transformation to the brightness of an image.

PHOTOMETRY

Original	Augmented

Original	Augmented

`aug = K.augmentation.RandomBrightness(brightness=(0.7, 1.3), p=1.0)`

19.

from kornia.augmentation

Shuffles the channels of a batch of multi-dimensional images.

PHOTOMETRY

Original	Augmented

Original	Augmented

`aug = K.augmentation.RandomChannelShuffle(p=1)`

20.

from kornia.augmentation

Applies a random transformation to the contrast of an image.

PHOTOMETRY

Original	Augmented

Original	Augmented

`aug = K.augmentation.RandomContrast(contrast=(0.8, 1.2), p=1.0)`

21.

from kornia.augmentation

Crops random patches of an image on a given size.

BASIC GEOMETRY

Original	Augmented

`aug = K.augmentation.RandomCrop(size=(1880, 1200), padding=(100, 0), fill=200, padding_mode='constant', p=1.0, cropping_mode='slice')`

22.

from kornia.augmentation

Implementation for CutMix: Regularization Strategy to Train Strong Classifiers with Localizable Features.

MIXED SAMPLE

Original	Augmented


`aug = K.augmentation.RandomCutMixV2() out = aug(images, labels)`

23.

from kornia.augmentation

Adds random elastic transformation to an image.

DISTORTION

Original	Augmented

`aug = K.augmentation.RandomElasticTransform(kernel_size=(63, 63), sigma=(32.0, 32.0), alpha=(1.0, 1.0), padding_mode='zeros', p=1)`

24.

from kornia.augmentation

Equalizes given image(s) randomly.

PHOTOMETRY

Original	Augmented

`aug = K.augmentation.RandomEqualize(p=1)`

25.

from kornia.augmentation

Erases a random rectangle of an image according to a probability p value.

DROPOUT

Original	Augmented

`aug = K.augmentation.RandomErasing(scale=(0.02, 0.33), ratio=(0.3, 3.3), value=0.0, p=1)`

26.

from kornia.augmentation

Adds random camera radial distortion.

DISTORTION

Original	Augmented

`center_x = torch.tensor([-.3, .3]) center_y = torch.tensor([-.3, .3]) gamma = torch.tensor([.9, 1.]) aug = K.augmentation.RandomFisheye(center_x, center_y, gamma, p=1)`

27.

from kornia.augmentation

Applies a random transformation to the gamma of an image.

PHOTOMETRY

Original	Augmented

`aug = K.augmentation.RandomGamma(gamma=(0.75, 0.75), gain=(1.0, 1.0), p=1.0)`

Original	Augmented

`aug = K.augmentation.RandomGamma(gamma=(0.75, 0.75), gain=(1.5, 1.5), p=1.0)`

Original	Augmented

`aug = K.augmentation.RandomGamma(gamma=(1.2, 1.2), gain=(1.5, 1.5), p=1.0)`

28.

from kornia.augmentation

Applies gaussian blur given image(s) randomly. The standard deviation is sampled for each instance.

BLUR

Original	Augmented

`aug = K.augmentation.RandomGaussianBlur(kernel_size=(11, 11), sigma=(2.5, 2.5), p=1)`

29.

from kornia.augmentation

Adds gaussian noise.

NOISE

Original	Augmented

`aug = K.augmentation.RandomGaussianNoise(mean=0.0, std=0.5, p=1)`

30.

from kornia.augmentation

Applies random transformation to Grayscale according to a probability p value.

PHOTOMETRY

Original	Augmented

`aug = K.augmentation.RandomGrayscale(p=1)`

31.

from kornia.augmentation

Applies a random horizontal flip to image(s) with a given probability.

BASIC GEOMETRY

Original	Augmented

`aug = K.augmentation.RandomHorizontalFlip(p=1)`

32.

from kornia.augmentation

Applies a random transformation to the hue of an image.

PHOTOMETRY

Original	Augmented

Original	Augmented

`aug = K.augmentation.RandomHue(hue=(-0.5, 0.5), p=1.0)`

33.

from kornia.augmentation

Inverts the images values randomly.

PHOTOMETRY

Original	Augmented

`aug = K.augmentation.RandomInvert(max_val=torch.tensor(1.0), p=1)`

34.

from kornia.augmentation

RandomJigsaw augmentation.

MIXED SAMPLE

Original	Augmented

`aug = K.augmentation.RandomJigsaw(grid=(2, 2), p=1)`

35.

from kornia.augmentation

Adds random blur with a median filter to an image.

BLUR

Original	Augmented

`aug = K.augmentation.RandomMedianBlur(kernel_size=(5, 5), p=1)`

36.

from kornia.augmentation

Implementation for mixup: BEYOND EMPIRICAL RISK MINIMIZATION.

MIXED SAMPLE

Original	Augmented


`aug = K.augmentation.RandomMixUpV2() out = aug(images, labels)`

37.

from kornia.augmentation

Mosaic augmentation.

MIXED SAMPLE

Original	Augmented


`aug = K.augmentation.RandomMosaic(p=1)`

38.

from kornia.augmentation

Performs motion blur on 2D images (4D tensor).

BLUR

Original	Augmented

`aug = K.augmentation.RandomMotionBlur(kernel_size=(15, 15), angle=(10, 10), direction=1.0, p=1)`

39.

from kornia.augmentation

Applies a random perspective transformation to an image with a given probability.

BASIC GEOMETRY

Original	Augmented

`aug = K.augmentation.RandomPerspective(distortion_scale=0.5, p=1, sampling_method='basic')`

40.

from kornia.augmentation

Applies planckian jitter transformation.

PHOTOMETRY

Original	Augmented

Original	Augmented

`aug = K.augmentation.RandomPlanckianJitter(mode='blackbody', select_from=None, p=1)`

41.

from kornia.augmentation

Adds brightness to the image based on a fractal map generated by the diamond square algorithm. Based on TorMentor: Deterministic dynamic-path, data augmentations with fractals.

Original	Augmented

Original	Augmented

`aug = K.augmentation.RandomPlasmaBrightness(roughness=(0.1, 0.7), intensity=(0.0, 1.0), p=1)`

42.

from kornia.augmentation

Adds contrast to the image based on a fractal map generated by the diamond square algorithm. Based on TorMentor: Deterministic dynamic-path, data augmentations with fractals.

Original	Augmented

Original	Augmented

`aug = K.augmentation.RandomPlasmaContrast(roughness=(0.1, 0.7), p=1)`

43.

from kornia.augmentation

Adds a particular shadow. Based on TorMentor: Deterministic dynamic-path, data augmentations with fractals.

Original	Augmented

Original	Augmented

`aug = K.augmentation.RandomPlasmaShadow(roughness=(0.1, 0.7), shade_intensity=(-1.0, 0.0), shade_quantity=(0.0, 1.0), p=1)`

44.

from kornia.augmentation

Posterizes given image(s) randomly.

STYLE FILTER

Original	Augmented

Original	Augmented

`aug = K.augmentation.RandomPosterize(bits=3, p=1)`

45.

from kornia.augmentation

Randomly shifts each channel of an image.

PHOTOMETRY

Original	Augmented

`aug = K.augmentation.RandomRGBShift(r_shift_limit=0.2, g_shift_limit=0.2, b_shift_limit=0.2, p=1)`

46.

from kornia.augmentation

Adds random rain to the image.

WEATHER

Original	Augmented

`aug = K.augmentation.RandomRain(p=1, number_of_drops=(20000, 20000), drop_height=(5, 20), drop_width=(-5, 5))`

47.

from kornia.augmentation

Crops random patches in an image and resizes to a given size.

BASIC GEOMETRY

Original	Augmented

`aug = K.augmentation.RandomResizedCrop(size=(1800, 1200), scale=(0.08, 1.0), ratio=(3.0 / 4.0, 4.0 / 3.0), p=1.0, cropping_mode='slice')`

48.

from kornia.augmentation

Applies a random rotation to image(s) given an amount of degrees.

BASIC GEOMETRY

Original	Augmented

`aug = K.augmentation.RandomRotation(degrees=(-15,-15), p=1)`

49.

from kornia.augmentation

Applies a random transformation to the saturation of an image.

PHOTOMETRY

Original	Augmented

Original	Augmented

`aug = K.augmentation.RandomSaturation(saturation=(0.5, 2), p=1.0)`

50.

from kornia.augmentation

Sharpens given image(s) randomly.

SHARPNESS

Original	Augmented

`aug = K.augmentation.RandomSharpness(sharpness=(7, 7), p=1)`

51.

from kornia.augmentation

Generates snow effect on given image(s).

WEATHER

Original	Augmented

`aug = K.augmentation.RandomSnow(snow_coefficient=(0.5, 0.5), brightness=(2, 2), p=1.0)`

52.

from kornia.augmentation

Solarizes given image(s) randomly.

PHOTOMETRY

Original	Augmented

Original	Augmented

`aug = K.augmentation.RandomSolarize(thresholds=0.1, additions=0.1, p=1)`

53.

from kornia.augmentation

Adds random noise to the Thin Plate Spline algorithm.

DISTORTION

Original	Augmented

Original	Augmented

`aug = K.augmentation.RandomThinPlateSpline(p=1)`

54.

from kornia.augmentation

Applies a random vertical flip to image(s) with a given probability.

BASIC GEOMETRY

Original	Augmented

`aug = K.augmentation.RandomVerticalFlip(p=1)`

55.

from kornia.augmentation

Resizes to size.

BASIC GEOMETRY

Original	Augmented

`aug = K.augmentation.Resize(size=(280), side='short', p=1.0)`

56.

from kornia.color

Applies the sepia filter to the image.

PHOTOMETRY

Original	Augmented

`# assuming input image as BGR aug = K.color.Sepia(rescale=False) out = aug(img) out = K.color.bgr_to_rgb(out)`

57.

from kornia.augmentation

Rescales an image so that minimum side is equal to max_size, keeping the aspect ratio of the initial image.

BASIC GEOMETRY

Original	Augmented

`aug = K.augmentation.SmallestMaxSize(max_size=300, p=1.0)`

58.

from kornia.augmentation.auto

Applies TrivialAugment augmentation strategies.

Original	Augmented

Original	Augmented

`aug = K.augmentation.auto.TrivialAugment(policy=None)`

59.

from kornia.filters

Creates an operator that sharpens image.

SHARPNESS

Original	Augmented

`aug = K.filters.UnsharpMask(kernel_size=(9, 9), sigma=(1.5, 1.5))`

60.

from kornia.enhance

61.

from kornia.morphology

Returns the closed image (erosion after a dilation), applying the same kernel in each channel.

MORPHOLOGY

Original	Augmented

`img = img[None, :] kernel = torch.ones(7, 7) out = K.morphology.closing(img, kernel)`

62.

from kornia.morphology

Returns the dilated image applying the same kernel in each channel.

MORPHOLOGY

Original	Augmented

`img = img[None, :] kernel = torch.ones(7, 7) out = K.morphology.dilation(img, kernel)`

63.

from kornia.morphology

Returns the eroded image applying the same kernel in each channel.

MORPHOLOGY

Original	Augmented

`img = img[None, :] kernel = torch.ones(7, 7) out = K.morphology.erosion(img, kernel)`

64.

from kornia.morphology

Returns the opened image (dilation after an erosion), applying the same kernel in each channel.

MORPHOLOGY

Original	Augmented

`img = img[None, :] kernel = torch.ones(7, 7) out = K.morphology.opening(img, kernel)`

Libraries

Kornia