7. SSD objects in Keras

Listing 11.7.1, displayed shortly, shows the SSD class. Two main routines are illustrated:

1. Creation of the SSD model using build_model()

2. Instantiating a data generator through build_generator()

build_model first creates a data dictionary from the train labels. The dictionary stores image filenames and ground truth bounding box coordinates and class for every object in each image. Afterward, the backbone and SSD network models are constructed. The most important product of model creation is self.ssd – the network model of SSD.

The labels are stored in a csv file. For the sample training images that is used in this book, the labels are saved in dataset/drinks/labels_train.csv with the format:

frame,xmin,xmax,ymin,ymax,class_id
0001000.jpg,310,445,104,443,1
0000999.jpg,194,354,96,478,1
0000998.jpg,105,383,134,244,1
0000997.jpg,157,493,89,194,1
0000996.jpg,51,435,207,347,1
0000995.jpg,183,536,156,283,1
0000994...