CNN Architecture

In this chapter, we will discuss an important class of deep learning network for images called convolutional neural networks (CNNs). The majority of the deep learning models built for image-related tasks, such as image recognition, classification, object detection, and so on, involve CNNs as their primary network. CNNs allow us to process the incoming data in a three-dimensional volume rather than a single dimension vector. Although CNNs are a class of neural networks (made up of weights, layers, and loss function), there are a lot of architectural differences to deep feedforward networks, which we will explain in this chapter. Just to give you an idea of how powerful CNNs can be, the ResNet CNN architecture achieved a top-error rate of 3.57% at the world famous image classification challenge—ILSVRC. This performance beats the human vision perception...