Auto Generated Documentation¶
The Augmentor image augmentation library.
Augmentor is a software package for augmenting image data. It provides a number of utilities that aid augmentation in a automated manner. The aim of the package is to make augmentation for machine learning tasks less prone to error, more reproducible, more efficient, and easier to perform.
Module by Module Documentation¶
Documentation of the Pipeline module¶
The Pipeline module is the user facing API for the Augmentor package. It
contains the Pipeline
class which is used to
create pipeline objects, which can be used to build an augmentation pipeline
by adding operations to the pipeline object.
For a good overview of how to use Augmentor, along with code samples and example images, can be seen in the Main Features section.

class
Augmentor.Pipeline.
DataFramePipeline
(source_dataframe, image_col, category_col, output_directory=u'output', save_format=None)[source]¶ Create a new Pipeline object pointing to dataframe containing the paths to your original image dataset.
Create a new Pipeline object, using the
source_dataframe
and the columnsimage_col
for the path of the image andcategory_col
for the name of the cateogryParameters:  source_dataframe – A Pandas DataFrame where the images are located
 output_directory – Specifies where augmented images should be saved to the disk. Default is the absolute path
 save_format – The file format to use when saving newly created, augmented images. Default is JPEG. Legal options are BMP, PNG, and GIF.
Returns: A
Pipeline
object.

class
Augmentor.Pipeline.
DataPipeline
(images, labels=None)[source]¶ The DataPipeline used to create augmented data that is not read from or saved to the hard disk. The class is provides beta functionality and will be incorporated into the standard Pipeline class at a later date.
Its main purpose is to provide functionality for augmenting images that have multiple masks.
See https://github.com/mdbloice/Augmentor/blob/master/notebooks/MultipleMaskAugmentation.ipynb for example usage.
DataPipeline objects are initialised by passing images and their corresponding masks (grouped as lists) along with an optional list of labels. If labels are provided, the augmented images and its corresponding label are returned, otherwise only the images are returned. Image data is returned in array format.
The images and masks that are passed can be of differing formats and have differing numbers of channels. For example, the ground truth data can be 3 channel RGB, while its mask images can be 1 channel monochrome.

augmentor_images
¶

labels
¶


class
Augmentor.Pipeline.
Pipeline
(source_directory=None, output_directory=u'output', save_format=None)[source]¶ The Pipeline class handles the creation of augmentation pipelines and the generation of augmented data by applying operations to this pipeline.
Create a new Pipeline object pointing to a directory containing your original image dataset.
Create a new Pipeline object, using the
source_directory
parameter as a source directory where your original images are stored. This folder will be scanned, and any valid file files will be collected and used as the original dataset that should be augmented. The scan will find any image files with the extensions JPEG/JPG, PNG, and GIF (case insensitive).Parameters:  source_directory – A directory on your filesystem where your original images are stored.
 output_directory – Specifies where augmented images should be saved to the disk. Default is the directory output relative to the path where the original image set was specified. If it does not exist it will be created.
 save_format – The file format to use when saving newly created, augmented images. Default is JPEG. Legal options are BMP, PNG, and GIF.
Returns: A
Pipeline
object.
add_further_directory
(new_source_directory, new_output_directory=u'output')[source]¶ Add a further directory containing images you wish to scan for augmentation.
Parameters:  new_source_directory (String) – The directory to scan for images.
 new_output_directory (String) – The directory to use for outputted, augmented images.
Returns: None

add_operation
(operation)[source]¶ Add an operation directly to the pipeline. Can be used to add custom operations to a pipeline.
To add custom operations to a pipeline, subclass from the Operation abstract base class, overload its methods, and insert the new object into the pipeline using this method.
See also
The
Operation
class.Parameters: operation (Operation) – An object of the operation you wish to add to the pipeline. Will accept custom operations written at runtime. Returns: None

black_and_white
(probability, threshold=128)[source]¶ Convert images to black and white. In other words convert the image to use a 1bit, binary palette. The threshold defaults to 128, but can be controlled using the
threshold
parameter.See also
The
greyscale()
function.Parameters:  probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed. For resizing, it is recommended that the probability be set to 1.
 threshold (Integer) – A value between 0 and 255 which controls the threshold point at which each pixel is converted to either black or white. Any values above this threshold are converted to white, and any values below this threshold are converted to black.
Returns: None

static
categorical_labels
(numerical_labels)[source]¶ Return categorical labels for an array of 0based numerical labels.
Parameters: numerical_labels (Arraylike list.) – The numerical labels. Returns: The categorical labels.

crop_by_size
(probability, width, height, centre=True)[source]¶ Crop an image according to a set of dimensions.
Crop each image according to
width
andheight
, by default at the centre of each image, otherwise at a random location within the image.See also
See
crop_random()
to crop a random, noncentred area of the image.If the crop area exceeds the size of the image, this function will crop the entire area of the image.
Parameters:  probability (Float) – The probability that the function will execute when the image is passed through the pipeline.
 width (Integer) – The width of the desired crop.
 height (Integer) – The height of the desired crop.
 centre (Boolean) – If True, crops from the centre of the image, otherwise crops at a random location within the image, maintaining the dimensions specified.
Returns: None

crop_centre
(probability, percentage_area, randomise_percentage_area=False)[source]¶ Crops the centre of an image as a percentage of the image’s area.
Parameters:  probability (Float) – The probability that the function will execute when the image is passed through the pipeline.
 percentage_area (Float) – The area, as a percentage of the current image’s area, to crop.
 randomise_percentage_area (Boolean) – If True, will use
percentage_area
as an upper bound and randomise the crop from between 0 andpercentage_area
.
Returns: None

crop_random
(probability, percentage_area, randomise_percentage_area=False)[source]¶ Crop a random area of an image, based on the percentage area to be returned.
This function crops a random area from an image, based on the area you specify using
percentage_area
.Parameters:  probability (Float) – The probability that the function will execute when the image is passed through the pipeline.
 percentage_area (Float) – The area, as a percentage of the current image’s area, to crop.
 randomise_percentage_area (Boolean) – If True, will use
percentage_area
as an upper bound and randomise the crop from between 0 andpercentage_area
.
Returns: None

flip_left_right
(probability)[source]¶ Flip (mirror) the image along its horizontal axis, i.e. from left to right.
See also
The
flip_top_bottom()
function.Parameters: probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed. Returns: None

flip_random
(probability)[source]¶ Flip (mirror) the image along either its horizontal or vertical axis.
This function mirrors the image along either the horizontal axis or the vertical access. The axis is selected randomly.
Parameters: probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed. Returns: None

flip_top_bottom
(probability)[source]¶ Flip (mirror) the image along its vertical axis, i.e. from top to bottom.
See also
The
flip_left_right()
function.Parameters: probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed. Returns: None

gaussian_distortion
(probability, grid_width, grid_height, magnitude, corner, method, mex=0.5, mey=0.5, sdx=0.05, sdy=0.05)[source]¶ Performs a random, elastic gaussian distortion on an image.
This function performs a randomised, elastic gaussian distortion controlled by the parameters specified. The grid width and height controls how fine the distortions are. Smaller sizes will result in larger, more pronounced, and less granular distortions. Larger numbers will result in finer, more granular distortions. The magnitude of the distortions can be controlled using magnitude. This can be random or fixed.
Good values for parameters are between 2 and 10 for the grid width and height, with a magnitude of between 1 and 10. Using values outside of these approximate ranges may result in unpredictable behaviour.
Parameters:  probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed.
 grid_width (Integer) – The number of rectangles in the grid’s horizontal axis.
 grid_height (Integer) – The number of rectangles in the grid’s vertical axis.
 magnitude (Integer) – The magnitude of the distortions.
 corner (String) – which corner of picture to distort. Possible values: “bell”(circular surface applied), “ul”(upper left), “ur”(upper right), “dl”(down left), “dr”(down right).
 method (String) – possible values: “in”(apply max magnitude to the chosen corner), “out”(inverse of method in).
 mex (Float) – used to generate 3d surface for similar distortions. Surface is based on normal distribution.
 mey (Float) – used to generate 3d surface for similar distortions. Surface is based on normal distribution.
 sdx (Float) – used to generate 3d surface for similar distortions. Surface is based on normal distribution.
 sdy (Float) – used to generate 3d surface for similar distortions. Surface is based on normal distribution.
Returns: None
For values
mex
,mey
,sdx
, andsdy
the surface is based on the normal distribution:\[e^{ \Big( \frac{(x\text{mex})^2}{\text{sdx}} + \frac{(y\text{mey})^2}{\text{sdy}} \Big) }\]

get_ground_truth_paths
()[source]¶ Returns a list of image and ground truth image path pairs. Used for verification purposes to ensure the ground truth images match to the images containing in the pipeline.
Returns: A list of tuples containing the image path and ground truth path pairs.

greyscale
(probability)[source]¶ Convert images to greyscale. For this operation, setting the
probability
to 1.0 is recommended.See also
The
black_and_white()
function.Parameters: probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed. For resizing, it is recommended that the probability be set to 1. Returns: None

ground_truth
(ground_truth_directory)[source]¶ Specifies a directory containing corresponding images that constitute respective ground truth images for the images in the current pipeline.
This function will search the directory specified by
ground_truth_directory
and will associate each ground truth image with the images in the pipeline by file name.Therefore, an image titled
cat321.jpg
will match with the imagecat321.jpg
in theground_truth_directory
. The function respects each image’s label, therefore the image namedcat321.jpg
with the labelcat
will match the imagecat321.jpg
in the subdirectorycat
relative toground_truth_directory
.Typically used to specify a set of ground truth or gold standard images that should be augmented alongside the original images of a dataset, such as image masks or semantic segmentation ground truth images.
A number of such data sets are openly available, see for example https://arxiv.org/pdf/1704.06857.pdf (GarciaGarcia et al., 2017).
Parameters: ground_truth_directory (String) – A directory containing the ground truth images that correspond to the images in the current pipeline. Returns: None.

histogram_equalisation
(probability=1.0)[source]¶ Apply histogram equalisation to the image.
Parameters: probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed. For histogram, equalisation it is recommended that the probability be set to 1. Returns: None

image_generator
()[source]¶ Deprecated. Use the sample function and return a generator. :return: A random image passed through the pipeline.

invert
(probability)[source]¶ Invert an image. For this operation, setting the
probability
to 1.0 is recommended.Warning
This function will cause errors if used on binary, 1bit palette images (e.g. black and white).
Parameters: probability – A value between 0 and 1 representing the probability that the operation should be performed. For resizing, it is recommended that the probability be set to 1. Returns: None

keras_generator
(batch_size, scaled=True, image_data_format=u'channels_last')[source]¶ Returns an image generator that will sample from the current pipeline indefinitely, as long as it is called.
Warning
This function returns images from the current pipeline with replacement.
You must configure the generator to provide data in the same format that Keras is configured for. You can use the functions
keras.backend.image_data_format()
andkeras.backend.set_image_data_format()
to get and set Keras’ image format at runtime.>>> from keras import backend as K >>> K.image_data_format() 'channels_first' >>> K.set_image_data_format('channels_last') >>> K.image_data_format() 'channels_last'
By default, Augmentor uses
'channels_last'
.Parameters:  batch_size (Integer) – The number of images to return per batch.
 scaled (Boolean) – True (default) if pixels are to be converted to float32 values between 0 and 1, or False if pixels should be integer values between 0255.
 image_data_format (String) – Either
'channels_last'
(default) or'channels_first'
.
Returns: An image generator.

keras_generator_from_array
(images, labels, batch_size, scaled=True, image_data_format=u'channels_last')[source]¶ Returns an image generator that will sample from the current pipeline indefinitely, as long as it is called.
Warning
This function returns images from
images
with replacement.You must configure the generator to provide data in the same format that Keras is configured for. You can use the functions
keras.backend.image_data_format()
andkeras.backend.set_image_data_format()
to get and set Keras’ image format at runtime.>>> from keras import backend as K >>> K.image_data_format() 'channels_first' >>> K.set_image_data_format('channels_last') >>> K.image_data_format() 'channels_last'
By default, Augmentor uses
'channels_last'
.Parameters:  images (Arraylike matrix. For greyscale images they can be
in the form
(l, x, y)
or(l, x, y, 1)
, wherel
is the number of images,x
is the image width andy
is the image height. For RGB/A images, the matrix should be in the form(l, x, y, n)
, wheren
is the number of layers, e.g. 3 for RGB or 4 for RGBA and CMYK.) – The images to augment using the current pipeline.  labels (List.) – The label associated with each image in
images
.  batch_size (Integer) – The number of images to return per batch.
 scaled (Boolean) – True (default) if pixels are to be converted to float32 values between 0 and 1, or False if pixels should be integer values between 0255.
 image_data_format – Either
'channels_last'
(default) or'channels_first'
. When'channels_last'
is specified the returned batch is in the form(batch_size, x, y, num_channels)
, while for'channels_last'
the batch is returned in the form(batch_size, num_channels, x, y)
.  image_data_format – String
Returns: An image generator.
 images (Arraylike matrix. For greyscale images they can be
in the form

keras_preprocess_func
()[source]¶ Returns the pipeline as a function that can be used with Keras ImageDataGenerator. The image array data fed to the returned function is supposed to have scaled to [0, 1]. It will be once converted to PIL format internally as Image.fromarray(np.uint8(255 * image)).
>>> import Augmentor >>> import torchvision >>> p = Augmentor.Pipeline() >>> p.rotate(probability=0.7, max_left_rotate=10, max_right_rotate=10) >>> p.zoom(probability=0.5, min_factor=1.1, max_factor=1.5) >>> from keras.preprocessing.image import ImageDataGenerator >>> datagen = ImageDataGenerator( >>> ... >>> preprocessing_function=p.keras_preprocess_func())
Returns: The pipeline as a function.

process
()[source]¶ This function is used to process every image in the pipeline exactly once.
This might be useful for resizing a dataset for example, and uses multithreading for fast execution.
It would make sense to set the probability of every operation in the pipeline to
1
when using this function.Returns: None

random_brightness
(probability, min_factor, max_factor)[source]¶ Random change brightness of an image.
Parameters:  probability – A value between 0 and 1 representing the probability that the operation should be performed.
 min_factor – The value between 0.0 and max_factor that define the minimum adjustment of image brightness. The value 0.0 gives a black image, value 1.0 gives the original image, value bigger than 1.0 gives more bright image.
 max_factor – A value should be bigger than min_factor that define the maximum adjustment of image brightness. The value 0.0 gives a black image, value 1.0 gives the original image, value bigger than 1.0 gives more bright image.
Returns: None

random_color
(probability, min_factor, max_factor)[source]¶ Random change saturation of an image.
Parameters:  probability – Controls the probability that the operation is performed when it is invoked in the pipeline.
 min_factor – The value between 0.0 and max_factor that define the minimum adjustment of image saturation. The value 0.0 gives a black and white image, value 1.0 gives the original image.
 max_factor – A value should be bigger than min_factor that define the maximum adjustment of image saturation. The value 0.0 gives a black and white image, value 1.0 gives the original image.
Returns: None

random_contrast
(probability, min_factor, max_factor)[source]¶ Random change image contrast.
Parameters:  probability – Controls the probability that the operation is performed when it is invoked in the pipeline.
 min_factor – The value between 0.0 and max_factor that define the minimum adjustment of image contrast. The value 0.0 gives s solid grey image, value 1.0 gives the original image.
 max_factor – A value should be bigger than min_factor that define the maximum adjustment of image contrast. The value 0.0 gives s solid grey image, value 1.0 gives the original image.
Returns: None

random_distortion
(probability, grid_width, grid_height, magnitude)[source]¶ Performs a random, elastic distortion on an image.
This function performs a randomised, elastic distortion controlled by the parameters specified. The grid width and height controls how fine the distortions are. Smaller sizes will result in larger, more pronounced, and less granular distortions. Larger numbers will result in finer, more granular distortions. The magnitude of the distortions can be controlled using magnitude. This can be random or fixed.
Good values for parameters are between 2 and 10 for the grid width and height, with a magnitude of between 1 and 10. Using values outside of these approximate ranges may result in unpredictable behaviour.
Parameters:  probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed.
 grid_width (Integer) – The number of rectangles in the grid’s horizontal axis.
 grid_height (Integer) – The number of rectangles in the grid’s vertical axis.
 magnitude (Integer) – The magnitude of the distortions.
Returns: None

random_erasing
(probability, rectangle_area)[source]¶ Work in progress. This operation performs a Random Erasing operation, as described in https://arxiv.org/abs/1708.04896 by Zhong et al.
Its purpose is to make models robust to occlusion, by randomly replacing rectangular regions with random pixel values.
For greyscale images the random pixels values will also be greyscale, and for RGB images the random pixels values will be in RGB.
This operation is subject to change, the original work describes several ways of filling the random regions, including a random solid colour or greyscale value. Currently this operations uses the method which yielded the best results in the tests performed by Zhong et al.
Parameters:  probability – A value between 0 and 1 representing the probability that the operation should be performed.
 rectangle_area – The percentage area of the image to occlude with the random rectangle, between 0.1 and 1.
Returns: None

remove_operation
(operation_index=1)[source]¶ Remove the operation specified by
operation_index
, if supplied, otherwise it will remove the latest operation added to the pipeline.See also
Use the
status()
function to find an operation’s index.Parameters: operation_index (Integer) – The index of the operation to remove. Returns: The removed operation. You can reinsert this at end of the pipeline using add_operation()
if required.

resize
(probability, width, height, resample_filter=u'BICUBIC')[source]¶ Resize an image according to a set of dimensions specified by the user in pixels.
Parameters:  probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed. For resizing, it is recommended that the probability be set to 1.
 width (Integer) – The new width that the image should be resized to.
 height (Integer) – The new height that the image should be resized to.
 resample_filter (String) – The resampling filter to use. Must be one of BICUBIC, BILINEAR, ANTIALIAS, or NEAREST.
Returns: None

rotate
(probability, max_left_rotation, max_right_rotation)[source]¶ Rotate an image by an arbitrary amount.
The operation will rotate an image by an random amount, within a range specified. The parameters
max_left_rotation
andmax_right_rotation
allow you to control this range. If you wish to rotate the images by an exact number of degrees, set bothmax_left_rotation
andmax_right_rotation
to the same value.Note
This function will rotate in place, and crop the largest possible rectangle from the rotated image.
In practice, angles larger than 25 degrees result in images that do not render correctly, therefore there is a limit of 25 degrees for this function.
If this function returns images that are not rendered correctly, then you must reduce the
max_left_rotation
andmax_right_rotation
arguments!Parameters:  probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed.
 max_left_rotation (Integer) – The maximum number of degrees the image can be rotated to the left.
 max_right_rotation (Integer) – The maximum number of degrees the image can be rotated to the right.
Returns: None

rotate180
(probability)[source]¶ Rotate an image by 180 degrees.
The operation will rotate an image by 180 degrees, and will be performed with a probability of that specified by the
probability
parameter.Parameters: probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed. Returns: None

rotate270
(probability)[source]¶ Rotate an image by 270 degrees.
The operation will rotate an image by 270 degrees, and will be performed with a probability of that specified by the
probability
parameter.Parameters: probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed. Returns: None

rotate90
(probability)[source]¶ Rotate an image by 90 degrees.
The operation will rotate an image by 90 degrees, and will be performed with a probability of that specified by the
probability
parameter.Parameters: probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed. Returns: None

rotate_random_90
(probability)[source]¶ Rotate an image by either 90, 180, or 270 degrees, selected randomly.
This function will rotate by either 90, 180, or 270 degrees. This is useful to avoid scenarios where images may be rotated back to their original positions (such as a
rotate90()
and arotate270()
being performed directly afterwards. The random rotation is chosen uniformly from 90, 180, or 270 degrees. The probability controls the chance of the operation being performed at all, and does not affect the rotation degree.Parameters: probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed. Returns: None

rotate_without_crop
(probability, max_left_rotation, max_right_rotation, expand=False, fillcolor=None)[source]¶ Rotate an image without automatically cropping.
The
expand
parameter controls whether the image is enlarged to contain the new rotated images, or if the image size is maintained Defaults tofalse
so that images maintain their dimensions when using this function.Parameters:  probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed.
 max_left_rotation (Integer) – The maximum number of degrees the image can be rotated to the left.
 max_right_rotation (Integer) – The maximum number of degrees the image can be rotated to the right.
 expand – Controls whether the image’s size should be
increased to accommodate the rotation. Defaults to
false
so that images maintain their original dimensions after rotation.  fillcolor – Specify color to fill points outside the boundaries of the input. Default value is None (points filled with black color). For example, in case of RGB color scheme simply use (r, g, b) tuple of int numbers.
Returns: None

sample
(n, multi_threaded=True)[source]¶ Generate
n
number of samples from the current pipeline.This function samples from the pipeline, using the original images defined during instantiation. All images generated by the pipeline are by default stored in an
output
directory, relative to the path defined during the pipeline’s instantiation.By default, Augmentor will use multithreading to increase the speed of processing the images. However, this may slow down some operations if the images are very small. Set
multi_threaded
toFalse
if slowdown is experienced.Parameters:  n (Integer) – The number of new samples to produce.
 multi_threaded (Boolean) – Whether to use multithreading to process the
images. Defaults to
True
.
Returns: None

sample_with_array
(image_array, save_to_disk=False)[source]¶ Generate images using a single image in arraylike format.
See also
See
keras_image_generator_without_replacement()
Parameters:  image_array – The image to pass through the pipeline.
 save_to_disk – Whether to save to disk or not (default).
Returns:

scale
(probability, scale_factor)[source]¶ Scale (enlarge) an image, while maintaining its aspect ratio. This returns an image with larger dimensions than the original image.
Use
resize()
to resize an image to absolute pixel values.Parameters:  probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed.
 scale_factor (Float) – The factor to scale by, which must be greater than 1.0.
Returns: None

set_save_format
(save_format)[source]¶ Set the save format for the pipeline. Pass the value
save_format="auto"
to allow Augmentor to choose the correct save format based on each individual image’s file extension.If
save_format
is set to, for example,save_format="JPEG"
orsave_format="JPG"
, Augmentor will attempt to save the files using the JPEG format, which may result in errors if the file cannot be saved in this format, such as trying to save PNG images with an alpha channel as JPEG.Parameters: save_format – The save format to save the images when writing to disk. Returns: None

static
set_seed
(seed)[source]¶ Set the seed of Python’s internal random number generator.
Parameters: seed (Integer) – The seed to use. Returns: None

shear
(probability, max_shear_left, max_shear_right)[source]¶ Shear the image by a specified number of degrees.
In practice, shear angles of more than 25 degrees can cause unpredictable behaviour. If you are observing images that are incorrectly rendered (e.g. they do not contain any information) then reduce the shear angles.
Parameters:  probability – The probability that the operation is performed.
 max_shear_left – The max number of degrees to shear to the left. Cannot be larger than 25 degrees.
 max_shear_right – The max number of degrees to shear to the right. Cannot be larger than 25 degrees.
Returns: None

skew
(probability, magnitude=1)[source]¶ Skew an image in a random direction, either left to right, top to bottom, or one of 8 corner directions.
To see examples of all the skew types, see Perspective Skewing.
Parameters:  probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed.
 magnitude (Float) – The maximum skew, which must be value between 0.1 and 1.0.
Returns: None

skew_corner
(probability, magnitude=1)[source]¶ Skew an image towards one corner, randomly by a random magnitude.
To see examples of the various skews, see Perspective Skewing.
Parameters:  probability – A value between 0 and 1 representing the probability that the operation should be performed.
 magnitude – The maximum skew, which must be value between 0.1 and 1.0.
Returns:

skew_left_right
(probability, magnitude=1)[source]¶ Skew an image by tilting it left or right by a random amount. The magnitude of this skew can be set to a maximum using the magnitude parameter. This can be either a scalar representing the maximum tilt, or vector representing a range.
To see examples of the various skews, see Perspective Skewing.
Parameters:  probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed.
 magnitude (Float) – The maximum tilt, which must be value between 0.1 and 1.0, where 1 represents a tilt of 45 degrees.
Returns: None

skew_tilt
(probability, magnitude=1)[source]¶ Skew an image by tilting in a random direction, either forwards, backwards, left, or right, by a random amount. The magnitude of this skew can be set to a maximum using the magnitude parameter. This can be either a scalar representing the maximum tilt, or vector representing a range.
To see examples of the various skews, see Perspective Skewing.
Parameters:  probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed.
 magnitude (Float) – The maximum tilt, which must be value between 0.1 and 1.0, where 1 represents a tilt of 45 degrees.
Returns: None

skew_top_bottom
(probability, magnitude=1)[source]¶ Skew an image by tilting it forwards or backwards by a random amount. The magnitude of this skew can be set to a maximum using the magnitude parameter. This can be either a scalar representing the maximum tilt, or vector representing a range.
To see examples of the various skews, see Perspective Skewing.
Parameters:  probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed.
 magnitude (Float) – The maximum tilt, which must be value between 0.1 and 1.0, where 1 represents a tilt of 45 degrees.
Returns: None

status
()[source]¶ Prints the status of the pipeline to the console. If you want to remove an operation, use the index shown and the
remove_operation()
method.See also
The
remove_operation()
function.See also
The
add_operation()
function.The status includes the number of operations currently attached to the pipeline, each operation’s parameters, the number of images in the pipeline, and a summary of the images’ properties, such as their dimensions and formats.
Returns: None

torch_transform
()[source]¶ Returns the pipeline as a function that can be used with torchvision.
>>> import Augmentor >>> import torchvision >>> p = Augmentor.Pipeline() >>> p.rotate(probability=0.7, max_left_rotate=10, max_right_rotate=10) >>> p.zoom(probability=0.5, min_factor=1.1, max_factor=1.5) >>> transforms = torchvision.transforms.Compose([ >>> p.torch_transform(), >>> torchvision.transforms.ToTensor(), >>> ])
Returns: The pipeline as a function.

zoom
(probability, min_factor, max_factor)[source]¶ Zoom in to an image, while maintaining its size. The amount by which the image is zoomed is a randomly chosen value between
min_factor
andmax_factor
.Typical values may be
min_factor=1.1
andmax_factor=1.5
.To zoom by a constant amount, set
min_factor
andmax_factor
to the same value.See also
See
zoom_random()
for zooming into random areas of the image.Parameters:  probability (Float) – A value between 0 and 1 representing the probability that the operation should be performed.
 min_factor (Float) – The minimum factor by which to zoom the image.
 max_factor (Float) – The maximum factor by which to zoom the image.
Returns: None

zoom_random
(probability, percentage_area, randomise_percentage_area=False)[source]¶ Zooms into an image at a random location within the image.
You can randomise the zoom level by setting the
randomise_percentage_area
argument to true.See also
See
zoom()
for zooming into the centre of images.Parameters:  probability – The probability that the function will execute when the image is passed through the pipeline.
 percentage_area – The area, as a percentage of the current image’s area, to crop.
 randomise_percentage_area – If True, will use
percentage_area
as an upper bound and randomise the crop from between 0 andpercentage_area
.
Returns: None
Documentation of the Operations module¶
The Operations module contains classes for all operations used by Augmentor.
The classes contained in this module are not called or instantiated directly
by the user, instead the user interacts with the
Pipeline
class and uses the utility functions contained
there.
In this module, each operation is a subclass of type Operation
.
The Pipeline
objects expect Operation
types, and therefore all operations are of type Operation
, and
provide their own implementation of the perform_operation()
function.
Hence, the documentation for this module is intended for developers who wish to extend Augmentor or wish to see how operations function internally.
For detailed information on extending Augmentor, see Extending Augmentor.

class
Augmentor.Operations.
BlackAndWhite
(probability, threshold)[source]¶ This class is used to convert images into black and white. In other words, into using a 1bit, monochrome binary colour palette. This is not to be confused with greyscale, where an 8bit greyscale pixel intensity range is used.
See also
The
Greyscale
class.As well as the required
probability
parameter, athreshold
can also be defined to define the cutoff point where a pixel is converted to black or white. Thethreshold
defaults to 128 at the userfacingblack_and_white()
function.Parameters:  probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline.
 threshold (Integer) – A value between 0 and 255 that defines the cut off point where an individual pixel is converted into black or white.

perform_operation
(images)[source]¶ Convert the image passed as an argument to black and white, 1bit monochrome. Uses the
threshold
passed to the constructor to control the cutoff point where a pixel is converted to black or white.Parameters: images (List containing PIL.Image object(s)) – The image to convert into monochrome. Returns: The transformed image(s) as a list of object(s) of type PIL.Image.

class
Augmentor.Operations.
Crop
(probability, width, height, centre)[source]¶ This class is used to crop images by absolute values passed as parameters.
As well as the always required
probability
parameter, the constructor requires awidth
to control the width of of the area to crop as well as aheight
parameter to control the height of the area to crop. Also, whether the area to crop should be taken from the centre of the image or from a random location within the image is toggled usingcentre
.Parameters:  probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline.
 width (Integer) – The width in pixels of the area to crop from the image.
 height (Integer) – The height in pixels of the area to crop from the image.
 centre (Boolean) – Whether to crop from the centre of the image or a random location within the image, while maintaining the size of the crop without cropping out of the original image’s area.

perform_operation
(images)[source]¶ Crop an area from an image, either from a random location or centred, using the dimensions supplied during instantiation.
Parameters: images (List containing PIL.Image object(s)) – The image(s) to crop the area from. Returns: The transformed image(s) as a list of object(s) of type PIL.Image.

class
Augmentor.Operations.
CropPercentage
(probability, percentage_area, centre, randomise_percentage_area)[source]¶ This class is used to crop images by a percentage of their area.
As well as the always required
probability
parameter, the constructor requires apercentage_area
to control the area of the image to crop in terms of its percentage of the original image, and acentre
parameter toggle whether a random area or the centre of the images should be cropped.Parameters:  probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline.
 percentage_area (Float) – The percentage area of the original image to crop. A value of 0.5 would crop an area that is 50% of the area of the original image’s size.
 centre (Boolean) – Whether to crop from the centre of the image or crop a random location within the image.

class
Augmentor.Operations.
CropRandom
(probability, percentage_area)[source]¶ Warning
This
CropRandom
class is currently not used by any of the userfacing functions in thePipeline
class.Parameters:  probability – Controls the probability that the operation is performed when it is invoked in the pipeline.
 percentage_area – The percentage area of the original image to crop. A value of 0.5 would crop an area that is 50% of the area of the original image’s size.

class
Augmentor.Operations.
Custom
(probability, custom_function, **function_arguments)[source]¶ Class that allows for a custom operation to be performed using Augmentor’s standard
Pipeline
object.Creates a custom operation that can be added to a pipeline.
To add a custom operation you can instantiate this class, passing a function pointer,
custom_function
, followed by an arbitrarily long list keyword arguments,**function_arguments
.See also
The
add_operation()
function.Parameters:  probability (Float) – The probability that the operation will be performed.
 custom_function (*Function) – The name of the function that performs your custom code. Must return an Image object and accept an Image object as its first parameter.
 function_arguments (dict) – The arguments for your custom operation’s code.

class
Augmentor.Operations.
Distort
(probability, grid_width, grid_height, magnitude)[source]¶ This class performs randomised, elastic distortions on images.
As well as the probability, the granularity of the distortions produced by this class can be controlled using the width and height of the overlaying distortion grid. The larger the height and width of the grid, the smaller the distortions. This means that larger grid sizes can result in finer, less severe distortions. As well as this, the magnitude of the distortions vectors can also be adjusted.
Parameters:  probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline.
 grid_width (Integer) – The width of the gird overlay, which is used by the class to apply the transformations to the image.
 grid_height (Integer) – The height of the gird overlay, which is used by the class to apply the transformations to the image.
 magnitude (Integer) – Controls the degree to which each distortion is applied to the overlaying distortion grid.

perform_operation
(images)[source]¶ Distorts the passed image(s) according to the parameters supplied during instantiation, returning the newly distorted image.
Parameters: images (List containing PIL.Image object(s)) – The image(s) to be distorted. Returns: The transformed image(s) as a list of object(s) of type PIL.Image.

class
Augmentor.Operations.
Flip
(probability, top_bottom_left_right)[source]¶ This class is used to mirror images through the x or y axes.
The class allows an image to be mirrored along either its x axis or its y axis, or randomly.
The direction of the flip, or whether it should be randomised, is controlled using the
top_bottom_left_right
parameter.Parameters:  probability – Controls the probability that the operation is performed when it is invoked in the pipeline.
 top_bottom_left_right –
Controls the direction the image should be mirrored. Must be one of
LEFT_RIGHT
,TOP_BOTTOM
, orRANDOM
.LEFT_RIGHT
defines that the image is mirrored along its x axis.TOP_BOTTOM
defines that the image is mirrored along its y axis.RANDOM
defines that the image is mirrored randomly along either the x or y axis.

perform_operation
(images)[source]¶ Mirror the image according to the attr:top_bottom_left_right` argument passed to the constructor and return the mirrored image.
Parameters: images (List containing PIL.Image object(s)) – The image(s) to mirror. Returns: The transformed image(s) as a list of object(s) of type PIL.Image.

class
Augmentor.Operations.
GaussianDistortion
(probability, grid_width, grid_height, magnitude, corner, method, mex, mey, sdx, sdy)[source]¶ This class performs randomised, elastic gaussian distortions on images.
As well as the probability, the granularity of the distortions produced by this class can be controlled using the width and height of the overlaying distortion grid. The larger the height and width of the grid, the smaller the distortions. This means that larger grid sizes can result in finer, less severe distortions. As well as this, the magnitude of the distortions vectors can also be adjusted.
Parameters:  probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline.
 grid_width (Integer) – The width of the gird overlay, which is used by the class to apply the transformations to the image.
 grid_height (Integer) – The height of the gird overlay, which is used by the class to apply the transformations to the image.
 magnitude (Integer) – Controls the degree to which each distortion is applied to the overlaying distortion grid.
 corner (String) – which corner of picture to distort. Possible values: “bell”(circular surface applied), “ul”(upper left), “ur”(upper right), “dl”(down left), “dr”(down right).
 method (String) – possible values: “in”(apply max magnitude to the chosen corner), “out”(inverse of method in).
 mex (Float) – used to generate 3d surface for similar distortions. Surface is based on normal distribution.
 mey (Float) – used to generate 3d surface for similar distortions. Surface is based on normal distribution.
 sdx (Float) – used to generate 3d surface for similar distortions. Surface is based on normal distribution.
 sdy (Float) – used to generate 3d surface for similar distortions. Surface is based on normal distribution.
For values
mex
,mey
,sdx
, andsdy
the surface is based on the normal distribution:\[e^{ \Big( \frac{(x\text{mex})^2}{\text{sdx}} + \frac{(y\text{mey})^2}{\text{sdy}} \Big) }\]
perform_operation
(images)[source]¶ Distorts the passed image(s) according to the parameters supplied during instantiation, returning the newly distorted image.
Parameters: images (List containing PIL.Image object(s)) – The image(s) to be distorted. Returns: The transformed image(s) as a list of object(s) of type PIL.Image.

class
Augmentor.Operations.
Greyscale
(probability)[source]¶ This class is used to convert images into greyscale. That is, it converts images into having only shades of grey (pixel value intensities) varying from 0 to 255 which represent black and white respectively.
See also
The
BlackAndWhite
class.As there are no further user definable parameters, the class is instantiated using only the
probability
argument.Parameters: probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline. 
perform_operation
(images)[source]¶ Converts the passed image to greyscale and returns the transformed image. There are no user definable parameters for this method.
Parameters: images (List containing PIL.Image object(s)) – The image to convert to greyscale. Returns: The transformed image(s) as a list of object(s) of type PIL.Image.


class
Augmentor.Operations.
HSVShifting
(probability, hue_shift, saturation_scale, saturation_shift, value_scale, value_shift)[source]¶ CURRENTLY NOT IMPLEMENTED.

class
Augmentor.Operations.
HistogramEqualisation
(probability)[source]¶ The class
HistogramEqualisation
is used to perform histogram equalisation on images passed to itsperform_operation()
function.As there are no further user definable parameters, the class is instantiated using only the
probability
argument.Parameters: probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline. 
perform_operation
(images)[source]¶ Performs histogram equalisation on the images passed as an argument and returns the equalised images. There are no user definable parameters for this method.
Parameters: images (List containing PIL.Image object(s)) – The image(s) on which to perform the histogram equalisation. Returns: The transformed image(s) as a list of object(s) of type PIL.Image.


class
Augmentor.Operations.
Invert
(probability)[source]¶ This class is used to negate images. That is to reverse the pixel values for any image processed by it.
As there are no further user definable parameters, the class is instantiated using only the
probability
argument.Parameters: probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline.

class
Augmentor.Operations.
Mixup
(probability, alpha=0.4)[source]¶ Implements the mixup augmentation method, as described in:
Zhang et al. (2018), mixup: Beyond Empirical Risk Minimization, arXiv:1710.09412
See http://arxiv.org/abs/1710.09412 for details.
Also see https://github.com/facebookresearch/mixupcifar10 for code which was followed to create this functionality in Augmentor.
The mixup augmentation technique trains a neural network on “combinations of pairs of examples and their labels” (Zhang et al., 2018).
In summary, mixup constructs training samples as follows:
\[\tilde{x} = \lambda x_i + (1\lambda)x_j\]\[\tilde{y} = \lambda y_i + (1\lambda)y_j\]where \((x_i, y_i)\) and \((x_j, y_j)\) are two samples from the training data, \(x_i, x_j\) are raw input vectors, \(y_i, y_j\) are onehot label encodings (such as [0.0, 1.0]), and \(\lambda \in [0, 1]\) where \(\lambda\) is sampled randomly from the Beta distribution, \(\beta(\alpha, \alpha)\).
The \(\alpha\) hyperparameter controls the strength of the interpolation between imagelabel pairs, where \(\alpha \in \{0, \infty \}\)
According to the paper referenced above, values for \(\alpha\) between 0.1 and 0.4 led to best performance. Smaller values for \(\alpha\) result in less mixup effect whereas larger values would tend to result in overfitting.
Note
Not yet enabled! This function is currently implemented but not enabled, as it requires each image’s label in order to operate  something which Augmentor was not designed to handle.
Parameters:  probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline.
 alpha (Float) – The alpha parameter controls the strength of the
interpolation between imagelabel pairs. It’s value can be any value
greater than 0. A smaller value for
alpha
results in more values closer to 0 or 1, meaning the mixup is more often closer to either of the images in the pair. Its value is set to 0.4 by default.

class
Augmentor.Operations.
Operation
(probability)[source]¶ The class
Operation
represents the base class for all operations that can be performed. Inherit fromOperation
, overload its methods, and instantiate super to create a new operation. See the section on extending Augmentor with custom operations at Extending Augmentor.All operations must at least have a
probability
which is initialised when creating the operation’s object.Parameters: probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline. 
perform_operation
(images)[source]¶ Perform the operation on the passed images. Each operation must at least have this function, which accepts a list containing objects of type PIL.Image, performs its operation, and returns a new list containing objects of type PIL.Image.
Parameters: images (List containing PIL.Image object(s)) – The image(s) to transform. Returns: The transformed image(s) as a list of object(s) of type PIL.Image.


class
Augmentor.Operations.
RandomBrightness
(probability, min_factor, max_factor)[source]¶ This class is used to random change image brightness.
required
probability
parameterrandom_brightness()
function.Parameters:  probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline.
 min_factor – The value between 0.0 and max_factor that define the minimum adjustment of image brightness. The value 0.0 gives a black image,The value 1.0 gives the original image, value bigger than 1.0 gives more bright image.
 max_factor (Float) – A value should be bigger than min_factor. that define the maximum adjustment of image brightness. The value 0.0 gives a black image, value 1.0 gives the original image, value bigger than 1.0 gives more bright image.

class
Augmentor.Operations.
RandomColor
(probability, min_factor, max_factor)[source]¶ This class is used to random change saturation of an image.
required
probability
parameterrandom_color()
function.Parameters:  probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline.
 min_factor – The value between 0.0 and max_factor that define the minimum adjustment of image saturation. The value 0.0 gives a black and white image, value 1.0 gives the original image.
 max_factor (Float) – A value should be bigger than min_factor. that define the maximum adjustment of image saturation. The value 0.0 gives a black and white image, value 1.0 gives the original image.

class
Augmentor.Operations.
RandomContrast
(probability, min_factor, max_factor)[source]¶ This class is used to random change contrast of an image.
required
probability
parameterrandom_contrast()
function.Parameters:  probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline.
 min_factor – The value between 0.0 and max_factor that define the minimum adjustment of image contrast. The value 0.0 gives s solid grey image, value 1.0 gives the original image.
 max_factor (Float) – A value should be bigger than min_factor. that define the maximum adjustment of image contrast. The value 0.0 gives s solid grey image, value 1.0 gives the original image.

class
Augmentor.Operations.
RandomErasing
(probability, rectangle_area)[source]¶ Class that performs Random Erasing, an augmentation technique described in https://arxiv.org/abs/1708.04896 by Zhong et al. To quote the authors, random erasing:
“… randomly selects a rectangle region in an image, and erases its pixels with random values.”
Exactly this is provided by this class.
Random Erasing can make a trained neural network more robust to occlusion.
The size of the random rectangle is controlled using the
rectangle_area
parameter. This area is random in its width and height.Parameters:  probability – The probability that the operation will be performed.
 rectangle_area – The percentage are of the image to occlude.

perform_operation
(images)[source]¶ Adds a random noise rectangle to a random area of the passed image, returning the original image with this rectangle superimposed.
Parameters: images (List containing PIL.Image object(s)) – The image(s) to add a random noise rectangle to. Returns: The transformed image(s) as a list of object(s) of type PIL.Image.

class
Augmentor.Operations.
Resize
(probability, width, height, resample_filter)[source]¶ This class is used to resize images by absolute values passed as parameters.
Accepts the required probability parameter as well as parameters to control the size of the transformed image.
Parameters:  probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline.
 width (Integer) – The width in pixels to resize the image to.
 height (Integer) – The height in pixels to resize the image to.
 resample_filter (String) – The resample filter to use. Must be one of
the standard PIL types, i.e.
NEAREST
,BICUBIC
,ANTIALIAS
, orBILINEAR
.

perform_operation
(images)[source]¶ Resize the passed image and returns the resized image. Uses the parameters passed to the constructor to resize the passed image.
Parameters: images (List containing PIL.Image object(s)) – The image to resize. Returns: The transformed image(s) as a list of object(s) of type PIL.Image.

class
Augmentor.Operations.
Rotate
(probability, rotation)[source]¶ This class is used to perform rotations on images in multiples of 90 degrees. Arbitrary rotations are handled by the
RotateRange
class.As well as the required
probability
parameter, therotation
parameter controls the rotation to perform, which must be one of90
,180
,270
or1
(see below).Parameters:  probability – Controls the probability that the operation is performed when it is invoked in the pipeline.
 rotation –
Controls the rotation to perform. Must be one of
90
,180
,270
or1
.90
rotate the image by 90 degrees.180
rotate the image by 180 degrees.270
rotate the image by 270 degrees.1
rotate the image randomly by either 90, 180, or 270 degrees.
See also
For arbitrary rotations, see the
RotateRange
class.

class
Augmentor.Operations.
RotateRange
(probability, max_left_rotation, max_right_rotation)[source]¶ This class is used to perform rotations on images by arbitrary numbers of degrees.
Images are rotated in place and an image of the same size is returned by this function. That is to say, that after a rotation has been performed, the largest possible area of the same aspect ratio of the original image is cropped from the skewed image, and this is then resized to match the original image size.
The method by which this is performed is described as follows:
\[E = \frac{\frac{\sin{\theta_{a}}}{\sin{\theta_{b}}}\Big(X\frac{\sin{\theta_{a}}}{\sin{\theta_{b}}} Y\Big)}{1\frac{(\sin{\theta_{a}})^2}{(\sin{\theta_{b}})^2}}\]which describes how \(E\) is derived, and then follows \(B = Y  E\) and \(A = \frac{\sin{\theta_{a}}}{\sin{\theta_{b}}} B\).
The Rotating section describes this in detail and has example images to demonstrate this.
As well as the required
probability
parameter, themax_left_rotation
parameter controls the maximum number of degrees by which to rotate to the left, while themax_right_rotation
controls the maximum number of degrees to rotate to the right.Parameters:  probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline.
 max_left_rotation (Integer) – The maximum number of degrees to rotate the image anticlockwise.
 max_right_rotation (Integer) – The maximum number of degrees to rotate the image clockwise.

perform_operation
(images)[source]¶ Perform the rotation on the passed
image
and return the transformed image. Uses themax_left_rotation
andmax_right_rotation
passed into the constructor to control the amount of degrees to rotate by. Whether the image is rotated clockwise or anticlockwise is chosen at random.Parameters: images (List containing PIL.Image object(s)) – The image(s) to rotate. Returns: The transformed image(s) as a list of object(s) of type PIL.Image.

class
Augmentor.Operations.
RotateStandard
(probability, max_left_rotation, max_right_rotation, expand=False, fillcolor=None)[source]¶ Class to perform rotations without automatically cropping the images, as opposed to the
RotateRange
class.See also
For arbitrary rotations with automatic cropping, see the
RotateRange
class.See also
For 90 degree rotations, see the
Rotate
class.Documentation to appear.

class
Augmentor.Operations.
Scale
(probability, scale_factor)[source]¶ This class is used to increase or decrease images in size by a certain factor, while maintaining the aspect ratio of the original image.
See also
The
Resize
class for resizing images by dimensions, and hence will not necessarily maintain the aspect ratio.This function will return images that are larger than the input images.
As the aspect ratio is always kept constant, only a
scale_factor
is required for scaling the image.Parameters:  probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline.
 scale_factor (Float) – The factor by which to scale, where 1.5 would result in an image scaled up by 150%.

class
Augmentor.Operations.
Shear
(probability, max_shear_left, max_shear_right)[source]¶ This class is used to shear images, that is to tilt them in a certain direction. Tilting can occur along either the x or yaxis and in both directions (i.e. left or right along the xaxis, up or down along the yaxis).
Images are sheared in place and an image of the same size as the input image is returned by this class. That is to say, that after a shear has been performed, the largest possible area of the same aspect ratio of the original image is cropped from the sheared image, and this is then resized to match the original image size. The Shearing section describes this in detail.
For sample code with image examples see Shearing.
The shearing is randomised in magnitude, from 0 to the
max_shear_left
or 0 tomax_shear_right
where the direction is randomised. The shear axis is also randomised i.e. if it shears up/down along the yaxis or left/right along the xaxis.Parameters:  probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline.
 max_shear_left (Integer) – The maximum shear to the left.
 max_shear_right (Integer) – The maximum shear to the right.

perform_operation
(images)[source]¶ Shears the passed image according to the parameters defined during instantiation, and returns the sheared image.
Parameters: images (List containing PIL.Image object(s)) – The image to shear. Returns: The transformed image(s) as a list of object(s) of type PIL.Image.

class
Augmentor.Operations.
Skew
(probability, skew_type, magnitude)[source]¶ This class is used to perform perspective skewing on images. It allows for skewing from a total of 12 different perspectives.
As well as the required
probability
parameter, the type of skew that is performed is controlled using askew_type
and amagnitude
parameter. Theskew_type
controls the direction of the skew, whilemagnitude
controls the degree to which the skew is performed.To see examples of the various skews, see Perspective Skewing.
Images are skewed in place and an image of the same size is returned by this function. That is to say, that after a skew has been performed, the largest possible area of the same aspect ratio of the original image is cropped from the skewed image, and this is then resized to match the original image size. The Perspective Skewing section describes this in detail.
Parameters:  probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline.
 skew_type (String) –
Must be one of
TILT
,TILT_TOP_BOTTOM
,TILT_LEFT_RIGHT
, orCORNER
.TILT
will randomly skew either left, right, up, or down. Left or right means it skews on the xaxis while up and down means that it skews on the yaxis.TILT_TOP_BOTTOM
will randomly skew up or down, or in other words skew along the yaxis.TILT_LEFT_RIGHT
will randomly skew left or right, or in other words skew along the xaxis.CORNER
will randomly skew one corner of the image either along the xaxis or yaxis. This means in one of 8 different directions, randomly.
To see examples of the various skews, see Perspective Skewing.
 magnitude (Integer) – The degree to which the image is skewed.

perform_operation
(images)[source]¶ Perform the skew on the passed image(s) and returns the transformed image(s). Uses the
skew_type
andmagnitude
parameters to control the type of skew to perform as well as the degree to which it is performed.If a list of images is passed, they must have identical dimensions. This is checked when we add the ground truth directory using
Pipeline.:func:`~Augmentor.Pipeline.Pipeline.ground_truth()
function.However, if this check fails, the skew function will be skipped and a warning thrown, in order to avoid an exception.
Parameters: images (List containing PIL.Image object(s)) – The image(s) to skew. Returns: The transformed image(s) as a list of object(s) of type PIL.Image.

class
Augmentor.Operations.
Zoom
(probability, min_factor, max_factor)[source]¶ This class is used to enlarge images (to zoom) but to return a cropped region of the zoomed image of the same size as the original image.
The amount of zoom applied is randomised, from between
min_factor
andmax_factor
. Set these both to the same value to always zoom by a constant factor.Parameters:  probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline.
 min_factor (Float) – The minimum amount of zoom to apply. Set both the
min_factor
andmin_factor
to the same values to zoom by a constant factor.  max_factor (Float) – The maximum amount of zoom to apply. Set both the
min_factor
andmin_factor
to the same values to zoom by a constant factor.

class
Augmentor.Operations.
ZoomGroundTruth
(probability, min_factor, max_factor)[source]¶ This class is used to enlarge images (to zoom) but to return a cropped region of the zoomed image of the same size as the original image.
The amount of zoom applied is randomised, from between
min_factor
andmax_factor
. Set these both to the same value to always zoom by a constant factor.Parameters:  probability (Float) – Controls the probability that the operation is performed when it is invoked in the pipeline.
 min_factor (Float) – The minimum amount of zoom to apply. Set both the
min_factor
andmin_factor
to the same values to zoom by a constant factor.  max_factor (Float) – The maximum amount of zoom to apply. Set both the
min_factor
andmin_factor
to the same values to zoom by a constant factor.

class
Augmentor.Operations.
ZoomRandom
(probability, percentage_area, randomise)[source]¶ This class is used to zoom into random areas of the image.
Zooms into a random area of the image, rather than the centre of the image, as is done by
Zoom
. The zoom factor is fixed unlessrandomise
is set toTrue
.Parameters:  probability – Controls the probability that the operation is performed when it is invoked in the pipeline.
 percentage_area – A value between 0.1 and 1 that represents the area that will be cropped, with 1 meaning the entire area of the image will be cropped and 0.1 mean 10% of the area of the image will be cropped, before zooming.
 randomise – If
True
, uses thepercentage_area
as an upper bound, and randomises the zoom level from between 0.1 andpercentage_area
.

perform_operation
(images)[source]¶ Randomly zoom into the passed
images
by first cropping the image based on thepercentage_area
argument, and then resizing the image to match the size of the input area.Effectively, you are zooming in on random areas of the image.
Parameters: images (List containing PIL.Image object(s)) – The image to crop an area from. Returns: The transformed image(s) as a list of object(s) of type PIL.Image.
Documentation of the ImageUtilities module¶
The ImageUtilities module provides a number of helper functions, as well as
the main AugmentorImage
class, that is used
throughout the package as a container class for images to be augmented.

class
Augmentor.ImageUtilities.
AugmentorImage
(image_path, output_directory, pil_images=None, array_images=None, path_images=None, class_label_int=None)[source]¶ Wrapper class containing paths to images, as well as a number of other parameters, that are used by the Pipeline and Operation modules to perform augmentation.
Each image that is found by Augmentor during the initialisation of a Pipeline object is contained with a new AugmentorImage object.
To initialise an AugmentorImage object for any image, the image’s file path is required, as well as that image’s output directory, which defines where any augmented images are stored.
Parameters:  image_path – The full path to an image.
 output_directory – The directory where augmented images for this image should be saved.

categorical_label
¶

class_label
¶

class_label_int
¶

file_format
¶

ground_truth
¶ The
ground_truth
property contains an absolute path to the ground truth file for an image.Getter: Returns this image’s ground truth file path. Setter: Sets this image’s ground truth file path. Type: String

image_arrays
¶

image_file_name
¶ The
image_file_name
property contains the file name of the image contained in this instance. There is no setter for this property.Getter: Returns this image’s file name. Type: String

image_path
¶ The
image_path
property contains the absolute file path to the image.Getter: Returns this image’s image path. Setter: Sets this image’s image path Type: String

label
¶

label_pair
¶

output_directory
¶ The
output_directory
property contains a path to the directory to which augmented images will be saved for this instance.Getter: Returns this image’s output directory. Setter: Sets this image’s output directory. Type: String

pil_images
¶

Augmentor.ImageUtilities.
extract_paths_and_extensions
(image_path)[source]¶ Extract an image’s file name, its extension, and its root path (the image’s absolute path without the file name).
Parameters: image_path (String) – The path to the image. Returns: A 3tuple containing the image’s file name, extension, and root path.

Augmentor.ImageUtilities.
scan_dataframe
(source_dataframe, image_col, category_col, output_directory)[source]¶

Augmentor.ImageUtilities.
scan_directory
(source_directory)[source]¶ Scan a directory for images, returning any images found with the extensions
.jpg
,.JPG
,.jpeg
,.JPEG
,.gif
,.GIF
,.img
,.IMG
,.png
,.PNG
,.tif
,.TIF
,.tiff
, or.TIFF
.Parameters: source_directory (String) – The directory to scan for images. Returns: A list of images found in the source_directory