yamle.third_party.imagenet_c package#

class yamle.third_party.imagenet_c.VisionCorruption(img_size, corruption_name)[source]#

Bases: object

Corrupts an image with a given corruption and severity.

The image must be a square and the size must be a multiple of 2.

Parameters:

  • img_size (Tuple[int, int, int]) – The size of the image. The image must be a square and the size must be a multiple of 2.

  • corruption_name (str) – The name of the corruption. Must be one of ‘gaussian_noise’, ‘shot_noise’, ‘impulse_noise’, ‘defocus_blur’, ‘glass_blur’, ‘motion_blur’, ‘zoom_blur’, ‘snow’, ‘frost’, ‘fog’, ‘brightness’, ‘contrast’, ‘elastic_transform’, ‘pixelate’, ‘jpeg_compression’, ‘speckle_noise’, ‘gaussian_blur’, ‘spatter’, ‘saturate’.

  • severity (int) – The severity of the corruption. Must be in [1, 6],

corruption_tuple = (<function gaussian_noise>, <function shot_noise>, <function impulse_noise>, <function defocus_blur>, <function glass_blur>, <function motion_blur>, <function zoom_blur>, <function snow>, <function frost>, <function fog>, <function brightness>, <function contrast>, <function elastic_transform>, <function pixelate>, <function jpeg_compression>, <function speckle_noise>, <function gaussian_blur>, <function spatter>, <function saturate>)#
corruption_dict = {'brightness': <function brightness>, 'contrast': <function contrast>, 'defocus_blur': <function defocus_blur>, 'elastic_transform': <function elastic_transform>, 'fog': <function fog>, 'frost': <function frost>, 'gaussian_blur': <function gaussian_blur>, 'gaussian_noise': <function gaussian_noise>, 'glass_blur': <function glass_blur>, 'impulse_noise': <function impulse_noise>, 'jpeg_compression': <function jpeg_compression>, 'motion_blur': <function motion_blur>, 'pixelate': <function pixelate>, 'saturate': <function saturate>, 'shot_noise': <function shot_noise>, 'snow': <function snow>, 'spatter': <function spatter>, 'speckle_noise': <function speckle_noise>, 'zoom_blur': <function zoom_blur>}#
corruption_names = ['gaussian_noise', 'shot_noise', 'impulse_noise', 'defocus_blur', 'glass_blur', 'motion_blur', 'zoom_blur', 'snow', 'frost', 'fog', 'brightness', 'contrast', 'elastic_transform', 'pixelate', 'jpeg_compression', 'speckle_noise', 'gaussian_blur', 'spatter', 'saturate']#
available_augmentations = ['gaussian_noise_0', 'gaussian_noise_1', 'gaussian_noise_2', 'gaussian_noise_3', 'gaussian_noise_4', 'shot_noise_0', 'shot_noise_1', 'shot_noise_2', 'shot_noise_3', 'shot_noise_4', 'impulse_noise_0', 'impulse_noise_1', 'impulse_noise_2', 'impulse_noise_3', 'impulse_noise_4', 'defocus_blur_0', 'defocus_blur_1', 'defocus_blur_2', 'defocus_blur_3', 'defocus_blur_4', 'glass_blur_0', 'glass_blur_1', 'glass_blur_2', 'glass_blur_3', 'glass_blur_4', 'motion_blur_0', 'motion_blur_1', 'motion_blur_2', 'motion_blur_3', 'motion_blur_4', 'zoom_blur_0', 'zoom_blur_1', 'zoom_blur_2', 'zoom_blur_3', 'zoom_blur_4', 'snow_0', 'snow_1', 'snow_2', 'snow_3', 'snow_4', 'frost_0', 'frost_1', 'frost_2', 'frost_3', 'frost_4', 'fog_0', 'fog_1', 'fog_2', 'fog_3', 'fog_4', 'brightness_0', 'brightness_1', 'brightness_2', 'brightness_3', 'brightness_4', 'contrast_0', 'contrast_1', 'contrast_2', 'contrast_3', 'contrast_4', 'elastic_transform_0', 'elastic_transform_1', 'elastic_transform_2', 'elastic_transform_3', 'elastic_transform_4', 'pixelate_0', 'pixelate_1', 'pixelate_2', 'pixelate_3', 'pixelate_4', 'jpeg_compression_0', 'jpeg_compression_1', 'jpeg_compression_2', 'jpeg_compression_3', 'jpeg_compression_4', 'speckle_noise_0', 'speckle_noise_1', 'speckle_noise_2', 'speckle_noise_3', 'speckle_noise_4', 'gaussian_blur_0', 'gaussian_blur_1', 'gaussian_blur_2', 'gaussian_blur_3', 'gaussian_blur_4', 'spatter_0', 'spatter_1', 'spatter_2', 'spatter_3', 'spatter_4', 'saturate_0', 'saturate_1', 'saturate_2', 'saturate_3', 'saturate_4']#
class yamle.third_party.imagenet_c.TabularCorruption(corruption_name)[source]#

Bases: object

Corrupts a tabular dataset with a given corruption and severity.

These augmentations are custom to this package and are not part of the original ImageNet-C corruptions.

Parameters:

  • corruption_name (str) – The name of the corruption. Must be one of ‘additive_gaussian_noise’, ‘multiplicative_gaussian_noise’, ‘additive_uniform_noise’, ‘multiplicative_uniform_noise’, ‘multiplicative_bernoulli_noise’.

  • severity (int) – The severity of the corruption. Must be in [1, 5],

corruption_tuple = (<function additive_gaussian_noise>, <function multiplicative_gaussian_noise>, <function additive_uniform_noise>, <function multiplicative_uniform_noise>, <function multiplicative_bernoulli_noise>)#
corruption_dict = {'additive_gaussian_noise': <function additive_gaussian_noise>, 'additive_uniform_noise': <function additive_uniform_noise>, 'multiplicative_bernoulli_noise': <function multiplicative_bernoulli_noise>, 'multiplicative_gaussian_noise': <function multiplicative_gaussian_noise>, 'multiplicative_uniform_noise': <function multiplicative_uniform_noise>}#
corruption_names = ['additive_gaussian_noise', 'multiplicative_gaussian_noise', 'additive_uniform_noise', 'multiplicative_uniform_noise', 'multiplicative_bernoulli_noise']#
available_augmentations = ['additive_gaussian_noise_0', 'additive_gaussian_noise_1', 'additive_gaussian_noise_2', 'additive_gaussian_noise_3', 'additive_gaussian_noise_4', 'multiplicative_gaussian_noise_0', 'multiplicative_gaussian_noise_1', 'multiplicative_gaussian_noise_2', 'multiplicative_gaussian_noise_3', 'multiplicative_gaussian_noise_4', 'additive_uniform_noise_0', 'additive_uniform_noise_1', 'additive_uniform_noise_2', 'additive_uniform_noise_3', 'additive_uniform_noise_4', 'multiplicative_uniform_noise_0', 'multiplicative_uniform_noise_1', 'multiplicative_uniform_noise_2', 'multiplicative_uniform_noise_3', 'multiplicative_uniform_noise_4', 'multiplicative_bernoulli_noise_0', 'multiplicative_bernoulli_noise_1', 'multiplicative_bernoulli_noise_2', 'multiplicative_bernoulli_noise_3', 'multiplicative_bernoulli_noise_4']#

Submodules#