dltk.io package

dltk.io.augmentation module

dltk.io.augmentation.add_gaussian_noise(image, sigma=0.05)

Add Gaussian noise to an image

Parameters:

• image (np.ndarray) – image to add noise to
• sigma (float) – stddev of the Gaussian distribution to generate noise from

Returns:

same as image but with added offset to each channel

Return type:

np.ndarray

dltk.io.augmentation.add_gaussian_offset(image, sigma=0.1)

Add Gaussian offset to an image. Adds the offset to each channel independently.

Parameters:

• image (np.ndarray) – image to add noise to
• sigma (float) – stddev of the Gaussian distribution to generate noise from

Returns:

same as image but with added offset to each channel

Return type:

np.ndarray

dltk.io.augmentation.elastic_transform(image, alpha, sigma)

Elastic deformation of images as described in [1].

[1] Simard, Steinkraus and Platt, “Best Practices for Convolutional

Neural Networks applied to Visual Document Analysis”, in Proc. of the International Conference on Document Analysis and Recognition, 2003.

Based on gist https://gist.github.com/erniejunior/601cdf56d2b424757de5

Parameters:

• image (np.ndarray) – image to be deformed
• alpha (list) – scale of transformation for each dimension, where larger values have more deformation
• sigma (list) – Gaussian window of deformation for each dimension, where smaller values have more localised deformation

Returns:

deformed image

Return type:

np.ndarray

dltk.io.augmentation.extract_class_balanced_example_array(image, label, example_size=[1, 64, 64], n_examples=1, classes=2, class_weights=None)

Extract training examples from an image (and corresponding label) subject

to class balancing. Returns an image example array and the corresponding label array.

Parameters:

• image (np.ndarray) – image to extract class-balanced patches from
• label (np.ndarray) – labels to use for balancing the classes
• example_size (list or tuple) – shape of the patches to extract
• n_examples (int) – number of patches to extract in total
• classes (int or list or tuple) – number of classes or list of classes to extract

Returns:

class-balanced patches extracted from full

images with the shape [batch, example_size..., image_channels]

Return type:

np.ndarray, np.ndarray

dltk.io.augmentation.extract_random_example_array(image_list, example_size=[1, 64, 64], n_examples=1)

Randomly extract training examples from image (and a corresponding label).

Returns an image example array and the corresponding label array.

Parameters:

• image_list (np.ndarray or list or tuple) – image(s) to extract random patches from
• example_size (list or tuple) – shape of the patches to extract
• n_examples (int) – number of patches to extract in total

Returns:

class-balanced patches extracted from full images with the shape [batch, example_size..., image_channels]

Return type:

np.ndarray, np.ndarray

dltk.io.augmentation.flip(imagelist, axis=1)

Randomly flip spatial dimensions

Parameters:

• imagelist (np.ndarray or list or tuple) – image(s) to be flipped
• axis (int) – axis along which to flip the images

Returns:

same as imagelist but randomly flipped

along axis

Return type:

np.ndarray or list or tuple

dltk.io.preprocessing module

dltk.io.preprocessing.normalise_one_one(image)

Image normalisation. Normalises image to fit [-1, 1] range.

dltk.io.preprocessing.normalise_zero_one(image)

Image normalisation. Normalises image to fit [0, 1] range.

dltk.io.preprocessing.resize_image_with_crop_or_pad(image, img_size=(64, 64, 64), **kwargs)

Image resizing. Resizes image by cropping or padding dimension

to fit specified size.

Parameters:

• image (np.ndarray) – image to be resized
• img_size (list or tuple) – new image size
• kwargs () – additional arguments to be passed to np.pad

Returns:

resized image

Return type:

np.ndarray

dltk.io.preprocessing.whitening(image)

Whitening. Normalises image to zero mean and unit variance.

dltk.core.io.abstract_reader module

class dltk.io.abstract_reader.IteratorInitializerHook

Bases: tensorflow.python.training.session_run_hook.SessionRunHook

Hook to initialise data iterator after Session is created.

after_create_session(session, coord)

Initialise the iterator after the session has been created.

class dltk.io.abstract_reader.Reader(read_fn, dtypes)

Bases: object

Wrapper for dataset generation given a read function

Constructs a Reader instance

Parameters:

• read_fn – Input function returning features which is a dictionary of string feature name to Tensor or SparseTensor. If it returns a tuple, first item is extracted as features. Prediction continues until input_fn raises an end-of-input exception (OutOfRangeError or StopIteration).
• dtypes – A nested structure of tf.DType objects corresponding to each component of an element yielded by generator.

get_dataset(file_references, mode, example_shapes=None, shuffle_cache_size=100, batch_size=4, params=None)

Function to get a tf.data.Dataset from a python generator.

Parameters:

• file_references – An array like structure that holds the reference to the file to read. It can also be None if not needed.
• mode – A tf.estimator.ModeKeys. It is passed on to read_fn to trigger specific functions there.
• example_shapes (optional) – A nested structure of lists or tuples corresponding to the shape of each component of an element yielded by generator.
• shuffle_cache_size (int, optional) – An int determining the number of examples that are held in the shuffle queue.
• batch_size (int, optional) – An int specifying the number of examples returned in a batch.
• params (dict, optional) – A dict passed on to the read_fn.

Returns:

The dataset holding the data to look at.

Return type:

tf.data.Dataset

get_inputs(file_references, mode, example_shapes=None, shuffle_cache_size=100, batch_size=4, params=None)

Function to provide the input_fn for a tf.Estimator.

Parameters:

• file_references – An array like structure that holds the reference to the file to read. It can also be None if not needed.
• mode – A tf.estimator.ModeKeys. It is passed on to read_fn to trigger specific functions there.
• example_shapes (optional) – A nested structure of lists or tuples corresponding to the shape of each component of an element yielded by generator.
• shuffle_cache_size (int, optional) – An int determining the number of examples that are held in the shuffle queue.
• batch_size (int, optional) – An int specifying the number of examples returned in a batch.
• params (dict, optional) – A dict passed on to the read_fn.

Returns:

a handle to the input_fn to be passed the relevant

tf estimator functions.

tf.train.SessionRunHook: A hook to initialize the queue within

the dataset.

Return type:

function

serving_input_receiver_fn(placeholder_shapes)

Build the serving inputs.

Parameters:placeholder_shapes – A nested structure of lists or tuples corresponding to the shape of each component of the feature elements yieled by the read_fn. Returns:A function to be passed to the tf.estimator.Estimator instance when exporting a saved model with estimator.export_savedmodel. Return type:function

Module contents