SSD is a very fast object detector that is well suited to be deployed on mobile and edge devices for real-time prediction. In this chapter, we will learn about how to develop a model using SSD and in the next chapter, we will evaluate its performance when deployed on edge devices. But before getting into the detail of SSD, we will get a quick overview of other object detector models we have learned about in this book so far.
We learned in Chapter 5, Neural Network Architecture and Models, that Faster R-CNN consists of 21,500 region proposals (60 x 40 sliding windows with 9 anchor boxes), which are warped into 2K fixed layers. These 2K layers are fed to a fully connected layer and bounding box regressors to detect the bounding boxes in an image. The 9 anchor boxes result from 3 scales with a box area of 1282, 2562, 5122, and three aspect ratios—1:1, 1...
