chapter three

3 Convolutional and ResNet Neural Networks

 

This chapter covers

  • Understanding the structure of convolutional neural networks (CNNs)
  • Constructing a ConvNet model
  • Designing and constructing a VGG model
  • Designing and constructing a Residual Network model

In chapter two, we introduced the fundamentals behind deep neural networks. The network architecture we covered was based on dense layers, which are also referred to as deep neural networks (DNN). We also demonstrated how to make a simple image classifier using dense layers, and discussed the limitations when attempting to scale an DNN to larger sizes of images. The introduction of constructing neural networks using convolutional layers for feature extraction and learning, known as convolutional neural networks (CNN), made it possible to scale image classifiers for practical applications.

In this chapter we will cover the design patterns and evolution in the design patterns for early state-of-the-art convolutional neural networks. We will cover three design patterns in this chapter, in sequence of their evolution:

  • ConvNet - Convolutional Network
  • VGG - Visual Graphics Group
  • Residual Network

3.1    Convolutional Neural Network (CNN)

3.1.1   Why we use a CNN over a DNN for image models

3.1.2   Downsampling (Resize)

3.1.3   Feature Detection

3.1.4   Pooling

3.1.5   Flattening

3.2    The ConvNet Design for a CNN

3.3    VGG Networks

3.4    ResNet Networks

3.4.1   Architecture

3.4.2   Batch Normalization

3.4.3   ResNet50

3.5    Summary