Convert coordinates for axis-aligned 2D boxes between two coordinate formats. Creates a copy of `tensor`, i.e. does not operate in place. Cu

Compute an array of the spatial positions and sizes of the anchor boxes for one particular classification layer of size `feature_map_

Compute the intersection-over-union similarity (also known as Jaccard similarity) of two axis-aligned 2D rectangular boxes or of multiple axi

Randomly change the brightness of the input image. Protected against overflow.

Flip the input image horizontally or vertically.

Scale the input image by a random factor picked from a uniform distribution over [min, max]. Returns: The scaled image, the asso

Randomly translate the input image horizontally and vertically. Arguments: image (array-like): The image to be translated. h

Perform histogram equalization on the input image. See https://en.wikipedia.org/wiki/Histogram_equalization.

The same greedy non-maximum suppression algorithm as above, but slightly modified for use as an internal function for per-class NMS in `decod

The same greedy non-maximum suppression algorithm as above, but slightly modified for use as an internal function in `decode_y2()`.

Convert ground truth bounding box data into a suitable format to train an SSD model. For each image in the batch, each ground truth

Produces an encoding template for the ground truth label tensor for a given batch. Note that all tensor creation, reshaping and conc

Compute the softmax log loss. Arguments: y_true (nD tensor): A TensorFlow tensor of any shape containing the ground trut

Compute smooth L1 loss, see references. Arguments: y_true (nD tensor): A TensorFlow tensor of any shape containing the g

Arguments: images_path (str): The filepath to the image samples. include_classes (list, optional): Either 'all' or a

All arguments need to be set to the same values as in the box encoding process, otherwise the behavior is undefined. Arguments:

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

Arguments: neg_pos_ratio (int, optional): The maximum number of negative (i.e. background) ground truth boxes to

Return an anchor box tensor based on the shape of the input tensor. The logic implemented here is identical to the logic in the modu

Compute the loss of the SSD model prediction against the ground truth. Arguments: y_true (array): A Numpy array of shape

A pure matrix multiplication implementation of `convert_coordinates()`. Although elegant, it turns out to be marginally slower on average th

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

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

Generate batches of samples and corresponding labels indefinitely from lists of filenames and labels. Returns two numpy arra

Returns: The list of filenames and the list of labels.

Returns: The number of image files in the initialized dataset.

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

Arguments: labels_path (str, optional): The filepath to a CSV file that contains one ground truth bounding box per line

This is a parser for the Pascal VOC datasets. It might be used for other datasets with minor changes to the code, but in its current

Perform offline image processing. This function the same image processing capabilities as the generator function above, but