Object localization is a widespread application of deep learning and has gained a lot of traction in the field of autonomous vehicles, facial detection, object tracking, and many more. Localizing an object is the identification of an area of interest in an image and encapsulating it with a bounding box. In Chapter 1, Understanding Neural Networks and Deep Neural Networks, and Chapter 2, Working with Convolutional Neural Networks, we worked on image classification, where the output of the network is the probability of each class. For this problem, we will use networks that are similar to the ones we used for image classification, except with a different set of target variables.

In object localization, we predict the output variables that represent the position of the object of interest in the entire input image. Using these, we draw bounding boxes on images. When object localization is accompanied by object classification, the technique is usually called object recognition. In this recipe, we are going to follow a simple generic method to localize a single object category in an image.