Regularizing a CNN with transfer learning for object detection

In this recipe, we will perform another typical task in computer vision – object detection. Before taking advantage of the power of transfer learning to help get better performances using a You Only Look Once (YOLO) model (a widely used class of models for object detection), we will give insights about what object detection is, the main methods and metrics, as well as the COCO dataset.

Object detection

Object detection is a computer vision task, involving both the identification and localization of objects of a given class (for example, a car, phone, person, or dog). As shown in Figure 10.14, the objects are usually localized, thanks to predicted bounding boxes, as well as predicted classes.

Figure 10.14 – An example of an image with object detection. Objects are detected with a bounding box and a class

Researchers have proposed many methods to help solve object detection...