ParticleDetection.modelling.augmentations

Collection of custom image augmentations extending the Detectron2 augmentation pool. These augmentations are intended to be used during the training process of a neural network using the Detectron2 framework.

Author: Adrian Niemann (adrian.niemann@ovgu.de)

Date: 31.10.2022

class GaussianBlur(*args: Any, **kwargs: Any)

Bases: Transform

Applies a Gaussian blur using the imgaug library.

Parameters

sigmas (Tuple[float, float]) –

Mean and variance of the constructed Gaussian kernel.

Default is (0.0, 2.0).

apply_coords(coords: ndarray)

apply_image(img: ndarray) → ndarray

inverse() → detectron2.data.transforms.Transform

class GaussianBlurAugmentation(*args: Any, **kwargs: Any)

Bases: Augmentation

Defines a strategy to blur images with a Gaussian blur operation.

Parameters

sigmas (Tuple[float, float]) –

Mean and variance of the constructed Gaussian kernel.

Default is (0.0, 2.0).

get_transform(*args) → detectron2.data.transforms.Transform

class Multiply(*args: Any, **kwargs: Any)

Bases: Transform

Applies a multipliction of each pixel with a certain value.

Parameters

mul (Tuple[float, float]) – The value with which to multiply the pixel values in each image. A value from the interval will be sampled per image and used for all pixels. Default is (0.8, 1.2)

apply_coords(coords: ndarray)

apply_image(img: ndarray)

inverse() → detectron2.data.transforms.Transform

class MultiplyAugmentation(*args: Any, **kwargs: Any)

Bases: Augmentation

Defines an augmentation strategy to multiply each pixel with a certain value.

Parameters

mul (Tuple[float, float]) –

The value with which to multiply the pixel values in each image. A value from the interval will be sampled per image and used for all pixels.

Default is (0.8, 1.2)

get_transform(*args) → detectron2.data.transforms.Transform

class Sharpen(*args: Any, **kwargs: Any)

Bases: Transform

Applies a sharpening transformation using the imgaug library.

Parameters

• alpha (Tuple[float, float]) –

  Blending factor of the sharpened image. A random value will be sampled from the interval for every image.

  Default is (0.0, 0.2).

• lightness (Tuple[float, float]) –

  Lightness/brightness of the sharped image. A random value will be sampled from the interval per image.

  Default is (0.8, 1.2).

apply_coords(coords: ndarray)

apply_image(img: ndarray) → ndarray

inverse() → detectron2.data.transforms.Transform

class SharpenAugmentation(*args: Any, **kwargs: Any)

Bases: Augmentation

Defines an augmentation strategy to sharpen images.

Parameters

• alpha (Tuple[float, float]) –

  Blending factor of the sharpened image. A random value will be sampled from the interval for every image.

  Default is (0.0, 0.2).

• lightness (Tuple[float, float]) –

  Lightness/brightness of the sharped image. A random value will be sampled from the interval per image.

  Default is (0.8, 1.2).

get_transform(*args) → detectron2.data.transforms.Transform

class SomeOf(*args: Any, **kwargs: Any)

Bases: AugmentationList

A list of Augmentations only some will be used of.

Randomly chooses Augmentation from the given list within the range of possible numbers of Augmentation ‘s.

Parameters

• augments (List[Augmentation]) – List of possible augmentations to choose from.

• lower (int) – Minimum amount of augmentations to choose.

• upper (int) – Maximum amount of augmentations to choose.

get_transform(*args) → detectron2.data.transforms.Transform