Practical Machine Learning for Computer Vision

Chapter 2. ML Models for Vision

In this chapter, you will learn how to represent images and train basic machine learning models to classify images. You will discover that the performance of linear and fully connected neural networks is poor on images. However, along the way, you will learn how to use the Keras API to implement ML primitives and train ML models.

Tip

The code for this chapter is in the 02_ml_models folder of the book’s GitHub repository. We will provide file names for code samples and notebooks where applicable.

A Dataset for Machine Perception

For the purposes of this book, it will be helpful if we take a single practical problem and build a variety of machine learning models to solve it. Assume that we have collected and labeled a dataset of nearly four thousand photographs of flowers. There are five types of flowers in the 5-flowers dataset (see Figure 2-1), and each image in the dataset has already been labeled with the type of flower it depicts.

Figure 2-1. The photographs in the 5-flowers dataset are of five types of flowers: daisies, dandelions, roses, sunflowers, and tulips.

Suppose we want to create a computer program that will, when provided an image, tell us what type of flower is in the image. We are asking the machine learning model to learn to perceive what’s in the image, so you might see this type of task called machine perception. Specifically, ...