augmentations

augmentations

Augmentation configuration module for data preprocessing.

This module defines configuration classes for various image augmentation strategies that can be instantiated at runtime using Hydra.

AugmentationConfig dataclass

AugmentationConfig(_target_=MISSING, p=1.0)

Base configuration for augmentation transforms.

Attributes:

Name	Type	Description
_target_	str	Import path instantiated by Hydra.
p	float	Probability of applying the transform.

ColorJitterConfig dataclass

ColorJitterConfig(_target_='albumentations.ColorJitter', p=0.5, brightness=0.3, contrast=0.4, saturation=0.5, hue=0.1)

Bases: AugmentationConfig

Random brightness, contrast, saturation, and hue jitter.

Attributes:

Name	Type	Description
_target_	str	Import path instantiated by Hydra.
brightness	float	Brightness jitter range.
contrast	float	Contrast jitter range.
saturation	float	Saturation jitter range.
hue	float	Hue jitter range.
p	float	Probability of applying the transform.

RandomSunFlareConfig dataclass

RandomSunFlareConfig(_target_='albumentations.RandomSunFlare', p=0.6, flare_roi=(0, 0, 1, 0.5), src_color=(255, 255, 255))

Bases: AugmentationConfig

Simulated sun-flare artifacts in the upper image region.

Attributes:

Name	Type	Description
_target_	str	Import path instantiated by Hydra.
flare_roi	tuple[float, float, float, float]	Region of the image where the flare can appear, as fractions.
src_color	tuple[int, int, int]	RGB color of the flare.
p	float	Probability of applying the transform.

RandomBrightnessContrastConfig dataclass

RandomBrightnessContrastConfig(_target_='albumentations.RandomBrightnessContrast', p=0.6, brightness_limit=0.4, contrast_limit=0.4)

Bases: AugmentationConfig

Random brightness and contrast shifts.

Attributes:

Name	Type	Description
_target_	str	Import path instantiated by Hydra.
brightness_limit	float	Maximum brightness shift as a fraction.
contrast_limit	float	Maximum contrast shift as a fraction.
p	float	Probability of applying the transform.

RandomGammaConfig dataclass

RandomGammaConfig(_target_='albumentations.RandomGamma', p=0.3, gamma_limit=(80, 120))

Bases: AugmentationConfig

Random gamma correction within the configured limits.

Attributes:

Name	Type	Description
_target_	str	Import path instantiated by Hydra.
gamma_limit	tuple[int, int]	Gamma correction range in percent.
p	float	Probability of applying the transform.

CLAHEConfig dataclass

CLAHEConfig(_target_='albumentations.CLAHE', p=0.3, clip_limit=4.0)

Bases: AugmentationConfig

Contrast-limited adaptive histogram equalization.

Attributes:

Name	Type	Description
_target_	str	Import path instantiated by Hydra.
clip_limit	float	Contrast-limiting threshold for histogram equalization.
p	float	Probability of applying the transform.

RandomShadowConfig dataclass

RandomShadowConfig(_target_='albumentations.RandomShadow', p=0.4)

Bases: AugmentationConfig

Random polygonal shadows cast over the image.

Attributes:

Name	Type	Description
_target_	str	Import path instantiated by Hydra.
p	float	Probability of applying the transform.

ImageCompressionConfig dataclass

ImageCompressionConfig(_target_='albumentations.ImageCompression', p=0.2, quality_range=(50, 100), compression_type='jpeg')

Bases: AugmentationConfig

Lossy compression artifacts within a random quality range.

Attributes:

Name	Type	Description
_target_	str	Import path instantiated by Hydra.
quality_range	tuple[int, int]	Inclusive (lower, upper) JPEG/WebP quality range.
compression_type	str	Compression codec, jpeg or webp.
p	float	Probability of applying the transform.

GaussianBlurConfig dataclass

GaussianBlurConfig(_target_='albumentations.GaussianBlur', p=0.5, blur_limit=(3, 7))

Bases: AugmentationConfig

Gaussian blur with a random kernel size.

Attributes:

Name	Type	Description
_target_	str	Import path instantiated by Hydra.
blur_limit	tuple[int, int]	Kernel size range for the blur.
p	float	Probability of applying the transform.

CoarseDropoutConfig dataclass

CoarseDropoutConfig(_target_='albumentations.CoarseDropout', p=0.3, num_holes_range=(8, 8), hole_height_range=(8, 8), hole_width_range=(8, 8))

Bases: AugmentationConfig

Random rectangular occlusion holes.

Attributes:

Name	Type	Description
_target_	str	Import path instantiated by Hydra.
num_holes_range	tuple[int, int]	Inclusive range for the number of dropped regions.
hole_height_range	tuple[int, int]	Inclusive height range of a dropped region in pixels.
hole_width_range	tuple[int, int]	Inclusive width range of a dropped region in pixels.
p	float	Probability of applying the transform.

ShiftScaleRotateConfig dataclass

ShiftScaleRotateConfig(_target_='albumentations.ShiftScaleRotate', p=0.5, rotate_limit=(0, 0), scale_limit=(-0.5, 0.6), shift_limit=(-0.0625, 0.0625))

Bases: AugmentationConfig

Random shift and scale; rotation stays disabled to keep kinematics consistent.

Attributes:

Name	Type	Description
_target_	str	Import path instantiated by Hydra.
rotate_limit	tuple[float, float]	Rotation range in degrees.
scale_limit	tuple[float, float]	Scaling range as a fraction.
shift_limit	tuple[float, float]	Translation range as a fraction of the image size.
p	float	Probability of applying the transform.

CenterCropConfig dataclass

CenterCropConfig(_target_='albumentations.CenterCrop', p=1.0, height=MISSING, width=MISSING)

Bases: AugmentationConfig

Center crop augmentation that preserves aspect ratio.

Attributes:

Name	Type	Description
_target_	str	Import path instantiated by Hydra.
height	int	Will be set from DataConfig.
width	int	Will be set from DataConfig.
p	float	Always apply if enabled.

RotateConfig dataclass

RotateConfig(_target_='albumentations.Rotate', p=0.5, limit=(-5, 5), interpolation=1)

Bases: AugmentationConfig

Rotation augmentation that requires special handling due to paired rotation of robot actions.

Attributes:

Name	Type	Description
_target_	str	Import path instantiated by Hydra.
limit	tuple[float, float]	Rotation range in degrees.
interpolation	int	cv2.INTER_LINEAR.
p	float	Probability of applying the transform.

AugmentationPipelineConfig dataclass

AugmentationPipelineConfig(_target_='albumentations.Compose', transforms=list())

Configuration for augmentation pipeline.

Attributes:

Name	Type	Description
_target_	str	Import path instantiated by Hydra.
transforms	list[Any]	Augmentation transforms applied in order.