mzbsuite module

Here we detail the main classes and functions in mzbsuite.

mzb_classification_dataloader

class mzbsuite.classification.mzb_classification_dataloader.Denormalize(mean, std)

De-normalizes an image given its mean and standard deviation.

class mzbsuite.classification.mzb_classification_dataloader.MZBLoader(dir_dict, ls_inds=[], learning_set='all', transforms=None, glob_pattern='*_rgb.*')

Class definition for the dataloader for the macrozoobenthos dataset.

Parameters:

• dir_dict (dict) – dictionary containing the paths to the folders of the dataset

• ls_inds (list) – indices of images to be used for the learning set, optional

• learning_set (str) – type of learning set to be used, optional, default: ‘all’

• transforms (torchvision.transforms) – list of transformations to apply to blobs. Optional, default: None

• glob_pattern (str) – glob pattern to use for finding images. Optional, default: ‘_rgb.’

static prepare_data(dir_dict: dict, ls_inds: list = [], glob_pattern: str = '*_rgb.*') → tuple

Prepare data for training and testing, returns image paths, labels and indices

Parameters:

• dir_dict (dict) – dictionary with keys as class names and values as paths to images

• ls_inds (list) – list of indices to be used for training or testing

• glob_pattern (str) – glob pattern to use for finding images

Returns:

img_paths – list of paths to images

Return type:

list

mzb_classification_pilmodel

mzb_skeletons_dataloader

class mzbsuite.skeletons.mzb_skeletons_dataloader.MZBLoader_skels(im_folder, bo_folder, he_folder, ls_inds=[], learning_set='all', transforms=None)

Class definition for the dataloader for the macrozoobenthos skeletons dataset.

Parameters:

• im_folder (Path) – folder path of input images

• bo_folder (Path) – folder path of body length skeleton masks

• he_folder (Path) – folder path of head length skeleton masks

• ls_inds (list) – indices of images to be used for the learning set, optional

• learning_set (str) – type of learning set to be used, optional, default: ‘all’

• transforms (torchvision.transforms) – optional, default: None

static prepare_data(im_folder, bo_folder=None, he_folder=None, ls_inds=[])

Prepares the data for the dataloader, loads it and returns it as numpy arrays.

mzb_skeletons_pilmodel

mzb_skeletons_helpers

class mzbsuite.skeletons.mzb_skeletons_helpers.Denormalize(mean, std)

Denormalize a tensor image with mean and standard deviation, for plotting purposes.

mzbsuite.skeletons.mzb_skeletons_helpers.get_endpoints(skeleton: ndarray) → List[Tuple[int, int]]

Given a skeletonised image, it will give the coordinates of the endpoints of the skeleton.

Parameters:

skeleton (numpy.ndarray) – The skeletonised image to detect the endpoints of

Returns:

endpoints – List of 2-tuples (x,y) containing the intersection coordinates

Return type:

list

mzbsuite.skeletons.mzb_skeletons_helpers.get_intersections(skeleton: ndarray) → List[Tuple[int, int]]

Given a skeletonised image, it will give the coordinates of the intersections of the skeleton.

Parameters:

skeleton (np.ndarray) – Binary image of the skeleton

Returns:

intersections – List of 2-tuples (x,y) containing the intersection coordinates

Return type:

list

mzbsuite.skeletons.mzb_skeletons_helpers.neighbours(x, y, image)

Return 8-neighbours of image point P1(x,y), in a clockwise order

Parameters:

• x (int) – x-coordinate of the point

• y (int) – y-coordinate of the point

• image (numpy.ndarray) – The image to find the neighbours of

Returns:

_ – List of 8-neighbours of the point in the image

Return type:

list

mzbsuite.skeletons.mzb_skeletons_helpers.paint_image(image: ndarray, mask: array, color: List[float]) → ndarray

Given an input image, a binary mask indicating where to paint, and a color to use, returns a new image where the pixels within the mask are colored with the specified color.

Parameters:

• (np.ndarray) (image) –

• (np.array) (mask) –

• (List[float]) (color) –

Returns:

rgb_fi (np.ndarray)

Return type:

New image with painted-in mask.

mzbsuite.skeletons.mzb_skeletons_helpers.paint_image_tensor(image: Tensor, masks: Tensor, color: List[float]) → Tensor

Given an input image, a binary mask indicating where to paint, and a color to use, returns a new image where the pixels within the mask are colored with the specified color.

Parameters:

• image (torch.Tensor) – Input image to paint.

• mask (torch.Tensor) – Binary mask indicating where to paint.

• color (List[float]) – List of 3 floats representing the RGB color to use.

Returns:

rgb_body – New image with painted pixels.

Return type:

torch.Tensor

mzbsuite.skeletons.mzb_skeletons_helpers.segment_skel(skeleton, inter, conn=1)

Custom function to segment a skeletonised image into individual branches. Each branch gets a unique ID.

Parameters:

• skeleton (numpy.ndarray) – The skeletonised image to segment

• inter (list) – List of 2-tuples (x,y) containing the intersection coordinates, as returned by the function find_intersections

• conn (int) – Connectivity of the skeleton. 1 for 4-connectivity, 2 for 8-connectivity

Returns:

• skel_labels (numpy.ndarray) – The labelled skeleton image

• edge_attributes (dict) – Dictionary containing the attributes of each edge (branch) (for now, its size in pixels)

• skprops (dict) – Dictionary containing the skimage.regionprops of each branch

mzbsuite.skeletons.mzb_skeletons_helpers.traverse_graph(graph: dict, init: int, end_nodes: List[int], debug: bool = False) → List[List[int]]

Function to traverse a graph from a starting node to a list of end nodes, and return all possible paths as a list of lists.

Parameters:

• graph (dict) – The graph to traverse

• init (int) – The starting node ID

• end_nodes (list) – List of end nodes

• debug (bool) – Whether to print debug information

Returns:

• all_paths (list) – List of lists containing all possible paths from init to end_nodes

• TODO

• Maybe.

• * Make it work for graphs with multiple paths between nodes and ensure that a subset of paths can be visited multiple times

• * Make a test for it

mzbsuite.utils