OpenCV 4 with Python Blueprints - Second Edition

Understanding object detectors

In Chapter 9, Learning to Classify and Localize Objects, we learned how to use the feature maps of a certain layer of a convolutional neural network to predict the bounding box of an object in the scene, which in our case was a head.

You might note that the difference between the localization network that we composed and the detection networks (that we used in this chapter) is that the detection networks predict multiple bounding boxes instead of a single one, as well as assigning a class to each of the bounding boxes.

Let's now make a smooth transition between the two architectures so that you can understand how object detection networks like YOLO and SSD work.