8.1.2.3. blueoil.data_augmentor

8.1.2.3.1. Module Contents

8.1.2.3.1.1. Classes

Blur

Gaussian blur filter.

Brightness

Adjust image brightness.

Color

Adjust image color.

Contrast

Adjust image contrast.

Crop

Crop image.

FlipLeftRight

Flip left right.

FlipTopBottom

Flip top bottom.

Hue

Change image hue.

Pad

Add padding to images.

RandomPatchCut

Cut out random patches of the image.

RandomErasing

Class that performs Random Erasing in Random Erasing Data Augmentation by Zhong et al.

RandomErasingForDetection

Class that performs Random Erasing in Random Erasing Data Augmentation by Zhong et al.

SSDRandomCrop

SSD random crop.

Rotate

Rotate.

8.1.2.3.1.2. Functions

_flip_left_right_boundingbox(image, boxes)

Flip left right only bounding box.

_flip_top_bottom_boundingbox(img, boxes)

Flip top bottom only bounding box.

_random_erasing_in_box(image, box, probability, sl, sh, r1, content_type, mean)

Random Erasing in a box (util func).

color_filter(img)

affine_scale(img, scale, fill_color=’white’)

Resize image to scale size keeping the aspect ratio and place it in center of fill color image.
class blueoil.data_augmentor.Blur(value=(0, 1))

Bases: blueoil.data_processor.Processor

Gaussian blur filter.

Reference:

https://pillow.readthedocs.io/en/stable/reference/ImageFilter.html#PIL.ImageFilter.GaussianBlur

Parameters

value (int | list | tuple) – Blur radius. Default is random number from 0 to 1. References default is 2.

__call__(self, image, **kwargs)
class blueoil.data_augmentor.Brightness(value=(0.75, 1.25))

Bases: blueoil.data_processor.Processor

Adjust image brightness.

Reference:

https://pillow.readthedocs.io/en/stable/reference/ImageEnhance.html#PIL.ImageEnhance.PIL.ImageEnhance.Brightness

Parameters

value (int | list | tuple) – An enhancement factor of 0.0 gives a black image. A factor of 1.0 gives the original image.

__call__(self, image, **kwargs)
class blueoil.data_augmentor.Color(value=(0.75, 1.25))

Bases: blueoil.data_processor.Processor

Adjust image color.

Reference:

https://pillow.readthedocs.io/en/stable/reference/ImageEnhance.html#PIL.ImageEnhance.PIL.ImageEnhance.Color

Parameters

value (int | list | tuple) – An enhancement factor of 0.0 gives a black and white image. A factor of 1.0 gives the original image.

__call__(self, image, **kwargs)
class blueoil.data_augmentor.Contrast(value=(0.75, 1.25))

Bases: blueoil.data_processor.Processor

Adjust image contrast.

Reference:

https://pillow.readthedocs.io/en/stable/reference/ImageEnhance.html#PIL.ImageEnhance.PIL.ImageEnhance.Contrast

Parameters

value (int | list | tuple) – An enhancement factor of 0.0 gives a solid grey image. A factor of 1.0 gives the original image.

__call__(self, image, **kwargs)
class blueoil.data_augmentor.Crop(size, resize=None)

Bases: blueoil.data_processor.Processor

Crop image.

Parameters

  • size (int | list | tuple) – Crop to this size.

  • resize (int | list | tuple) – If there are resize param, resize and crop.

__call__(self, image, mask=None, **kwargs)
blueoil.data_augmentor._flip_left_right_boundingbox(image, boxes)

Flip left right only bounding box.

Parameters

  • image (np.ndarray) – a image. shape is [height, width, channel]

  • boxes (np.ndarray) – bounding boxes. shape is [num_boxes, 5(x, y, w, h, class_id)]

class blueoil.data_augmentor.FlipLeftRight(probability=0.5)

Bases: blueoil.data_processor.Processor

Flip left right.

Parameters

probability (number) – Probability for flipping.

__call__(self, image, mask=None, gt_boxes=None, **kwargs)
blueoil.data_augmentor._flip_top_bottom_boundingbox(img, boxes)

Flip top bottom only bounding box.

Parameters

  • img – np array image.

  • boxes (np.ndarray) – bounding boxes. shape is [num_boxes, 5(x, y, w, h, class_id)]

class blueoil.data_augmentor.FlipTopBottom(probability=0.5)

Bases: blueoil.data_processor.Processor

Flip top bottom.

Parameters

probability (number) – Probability for flipping.

__call__(self, image, mask=None, gt_boxes=None, **kwargs)
Parameters

  • image (np.ndarray) – a image. shape is [height, width, channel]

  • mask (np.ndarray) – Annotation data for segmentation. shape is [height, width] or [height, width, channel]

  • gt_boxes – Ground truth boxes in the image. shape is [num_boxes, 5(x, y, w, h, class)]

Returns

Contains processed image, mask, gt_boxes and etc.

Return type

all args (dict)

class blueoil.data_augmentor.Hue(value=(-10, 10))

Bases: blueoil.data_processor.Processor

Change image hue.

Parameters

value (int | list | tuple) – Assume the value in -255, 255. When the value is 0, nothing to do.

__call__(self, image, **kwargs)
class blueoil.data_augmentor.Pad(value, fill=0)

Bases: blueoil.data_processor.Processor

Add padding to images.

Parameters

  • value (int or tuple) – Padding on each border. If a single int is provided this is used to pad all borders. If tuple of length 2 is provided this is the padding on left/right and top/bottom respectively. If a tuple of length 4 is provided this is the padding for the left, top, right and bottom borders respectively.

  • fill (int) – Pixel fill value. Default is 0.

__call__(self, image, mask=None, **kwargs)
Parameters

  • image (np.ndarray) – a image. shape is [height, width, channel]

  • mask (np.ndarray) – Annotation data for segmentation. shape is [height, width] or [height, width, channel]

Returns

Contains processed image, mask and etc.

Return type

all args (dict)

class blueoil.data_augmentor.RandomPatchCut(num_patch=1, max_size=10, square=True)

Bases: blueoil.data_processor.Processor

Cut out random patches of the image.

Parameters

  • num_patch (int) – number of random patch cut-outs to generate

  • max_size (int) – maximum size of the patch edge, in percentages of image size

  • square (bool) – force square aspect ratio for patch shape

__call__(self, image, **kwargs)

Cut random patches.

Parameters

image (np.ndarray) – a image. shape is [height, width, channel]

Returns

Contains processed image and etc.

Return type

all args (dict)

class blueoil.data_augmentor.RandomErasing(probability=0.5, sl=0.02, sh=0.4, r1=0.3, content_type='mean', mean=[125, 122, 114])

Bases: blueoil.data_processor.Processor

Class that performs Random Erasing in Random Erasing Data Augmentation by Zhong et al. The following Args(hyper parameters) are referred to the paper.

Parameters

  • probability (float) – The probability that the operation will be performed.

  • sl (float) – min erasing area

  • sh (float) – max erasing area

  • r1 (float) – min aspect ratio

  • content_type (string) – type of erasing value: {“mean”, “random”}

  • mean (list) – erasing value if you use “mean” mode (mean ImageNet pixel value)

__call__(self, image, **kwargs)

Random Erasing in an entire image.

Parameters

  • image (np.ndarray) – a image. shape is [height, width, channel]

  • gt_boxes – Ground truth boxes in the image. shape is [num_boxes, 5(x, y, w, h, class)]

Returns

Contains processed image and etc.

Return type

all args (dict)

class blueoil.data_augmentor.RandomErasingForDetection(probability=0.5, sl=0.02, sh=0.2, r1=0.3, content_type='mean', mean=[125, 122, 114], i_a=True, o_a=True)

Bases: blueoil.data_processor.Processor

Class that performs Random Erasing in Random Erasing Data Augmentation by Zhong et al. The following Args(hyper parameters) are referred to the paper. This Augmentation can be used when you train object detection task.

Parameters

  • probability (float) – The probability that the operation will be performed.

  • sl (float) – min erasing area

  • sh (float) – max erasing area

  • r1 (float) – min aspect ratio

  • content_type (string) – type of erasing value: {“mean”, “random”}

  • mean (list) – erasing value if you use “mean” mode (mean ImageNet pixel value)

  • i_a (bool) – image-aware, random erase an entire image.

  • o_a (bool) – object-aware, random erase each object bounding boxes.

__call__(self, image, gt_boxes=None, **kwargs)

Random Erasing in boxes (and an entire image).

Parameters

  • image (np.ndarray) – a image. shape is [height, width, channel]

  • gt_boxes – Ground truth boxes in the image. shape is [num_boxes, 5(x, y, w, h, class)]

Returns

Contains processed image and etc.

Return type

all args (dict)

blueoil.data_augmentor._random_erasing_in_box(image, box, probability, sl, sh, r1, content_type, mean)

Random Erasing in a box (util func).

Parameters

  • image (np.ndarray) – a image. shape is [height, width, channel]

  • box – Ground truth boxes in the image. shape is boxes, [5(x, y, w, h, class)]

class blueoil.data_augmentor.SSDRandomCrop(min_crop_ratio=0.3)

Bases: blueoil.data_processor.Processor

SSD random crop. :param min_crop_ratio: Minimum crop ratio for cropping the :type min_crop_ratio: number

__call__(self, image, gt_boxes, **kwargs)

SSDRandomCrop

Parameters

  • image (np.ndarray) – a image. shape is [height, width, channel]

  • gt_boxes – Ground truth boxes in the image. shape is [num_boxes, 5(x, y, w, h, class)]

Returns

Contains processed image, gt_boxes and etc.

Return type

all args (dict)

class blueoil.data_augmentor.Rotate(angle_range=(0, 90))

Bases: blueoil.data_processor.Processor

Rotate.

Parameters

angle_range (int | list | tuple) – Angle range.

__call__(self, image, mask=None, **kwargs)
Parameters

  • image (np.ndarray) – a image. shape is [height, width, channel]

  • mask (np.ndarray) – Annotation data for segmentation. shape is [height, width] or [height, width, channel]

Returns

Contains processed image, mask and etc.

Return type

all args (dict)

blueoil.data_augmentor.color_filter(img)
blueoil.data_augmentor.affine_scale(img, scale, fill_color='white')

Resize image to scale size keeping the aspect ratio and place it in center of fill color image.