Returns: An item of the sample space, i.e. a 2-tuple of scalars.

Arguments: labels (array): The labels to be filtered. This is an array with shape `(m,n)`, where `m` is the numbe

Arguments: labels (array): The labels to be tested. The box coordinates are expected to be in the image's coordin

Converts ground truth bounding box data into a suitable format to train an SSD model. Arguments: ground_truth_labels (

Computes the mean average precision of the given Keras SSD model on the given dataset. Optionally also returns the averages precisi

Returns: A 4-tuple `(ymin, xmin, height, width)` that represents the coordinates of the generated patch.

Initializes the data generator. You can either load a dataset directly here in the constructor, e.g. an HDF5 dataset, or you can us

Arguments: current (str, optional): The current color space of the images. Can be one of 'RGB' and 'HSV'.

Arguments: to (string, optional): To which datatype to convert the input images. Can be either of 'uint8' and 'fl

Arguments: delta (int): An integer in the closed interval `[-180, 180]` that determines the hue change, where a c

Arguments: max_delta (int): An integer in the closed interval `[0, 180]` that determines the maximal absolute hue

Arguments: factor (float): A float greater than zero that determines saturation change, where values less than on

Arguments: lower (float, optional): A float greater than zero, the lower bound for the random saturation change.

Arguments: delta (int): An integer, the amount to add to or subtract from the intensity of every pixel.

Arguments: lower (int, optional): An integer, the lower bound for the random brightness change. upper (int, optional)

Arguments: factor (float): A float greater than zero that determines contrast change, where values less than one

Arguments: lower (float, optional): A float greater than zero, the lower bound for the random contrast change.

Arguments: gamma (float): A float greater than zero that determines gamma change.

Arguments: lower (float, optional): A float greater than zero, the lower bound for the random gamma change.

Arguments: prob (float, optional): `(1 - prob)` determines the probability with which the original, unaltered ima

Arguments: order (tuple): A tuple of integers that defines the desired channel order of the input images after th

Arguments: prob (float, optional): `(1 - prob)` determines the probability with which the original, unaltered ima

Arguments: sample_space (list or tuple): A list, tuple, or array-like object of shape `(n, 2)` that contains `n`

Arguments: check_overlap (bool, optional): Whether or not to enforce the overlap requirements defined by `overlap

Arguments: overlap_criterion (str, optional): Can be either of 'center_point', 'iou', or 'area'. Determines which

Arguments: height (int): The desired height of the output images in pixels. width (int): The desired width of the out

Arguments: height (int): The desired height of the output image in pixels. width (int): The desired width of the outp

Arguments: dim (str, optional): Can be either of 'horizontal' and 'vertical'. If 'horizontal', images will be fli

Arguments: dim (str, optional): Can be either of 'horizontal' and 'vertical'. If 'horizontal', images will be fli

Arguments: dy (float): The fraction of the image height by which to translate images along the vertical axis. Pos

Arguments: dy_minmax (list/tuple, optional): A 2-tuple `(min, max)` of non-negative floats that determines the mi

Arguments: factor (float): The fraction of the image size by which to scale images. Must be positive. clip_boxes (boo

Arguments: min_factor (float, optional): The minimum fraction of the image size by which to scale images. Must be

Arguments: angle (int): The angle in degrees by which to rotate the images counter-clockwise. Only 90, 180, and 2

Arguments: angle (list): The list of angles in degrees from which one is randomly selected to rotate the images c

Arguments: img_height (int): The height of the input images. img_width (int): The width of the input images.

Arguments: model (Keras model): A Keras SSD model object. n_classes (int): The number of positive classes, e.g. 20

Arguments: img_height (int): The height of the image for which the patch coordinates shall be generated. Doesn't

Arguments: patch_ymin (int, optional): The vertical coordinate of the top left corner of the output patch relativ

Arguments: patch_coord_generator (PatchCoordinateGenerator): A `PatchCoordinateGenerator` object to generate the

Arguments: patch_coord_generator (PatchCoordinateGenerator): A `PatchCoordinateGenerator` object to generate the

Arguments: patch_aspect_ratio (float): The fixed aspect ratio that all sampled patches will have. box_filter (BoxFilt

Arguments: patch_aspect_ratio (float): The fixed aspect ratio that all sampled patches will have. background (list/tu

Compose arbitrarily many functions, evaluated left to right. Reference: https://mathieularose.com/function-composition-in-python/

A matrix multiplication implementation of `convert_coordinates()`. Supports only conversion between the 'centroids' and 'minmax' formats.

Converts the currently loaded dataset into a HDF5 file. This HDF5 file contains all images as uncompressed arrays in a contiguous b

This decoder performs the same processing as `decode_detections()`, but the output format for each left-over predicted box is `[box_id, cla

Convert model prediction output back to a format that contains only the positive box predictions (i.e. the same format that `enconde_y()` t

Returns: 4-tuple containing lists and/or `None` for the filenames, labels, image IDs, and evaluation-neutrality an

Returns a list of the number of boxes that each predictor layer predicts. `aspect_ratios` must be a nested list, containing a list of aspec

For a given prediction tensor decoded with `decode_detections_debug()`, returns a list with the indices of the predictor layers that made e

random preprocessing for real-time data augmentation

Perform greedy non-maximum suppression on the input boxes. Greedy NMS works by selecting the box with the highest score and removing a

Computes the intersection areas of two sets of axis-aligned 2D rectangular boxes. Let `boxes1` and `boxes2` contain `m` and `n` boxes, resp