Throughout the chapters in this book, we have seen various machine learning models, each progressively increasing their perceptual abilities. By this, I mean that we were first introduced to a model capable of classifying a single object present in an image. Then came a model that was able to classify not only multiple objects but also their corresponding bounding boxes. In this chapter, we continue this progression by introducing semantic segmentation, in other words, being able to assign each pixel to a specific class, as shown in the following figure: 

This allows for a greater understanding of the scene and, therefore, opportunities for more intelligible interfaces and services. But this is not the main focus of this chapter. In this chapter, we will use semantic segmentation to create an image effects application as a way to demonstrate imperfect predictions. We'll be using this to motivate a discussion on one of the most important aspects of designing and building machine learning (or artificial intelligence) interfaces—dealing with probabilistic, or imperfect, outcomes from models.

By the end of this chapter, you will have:

• An understanding semantic of segmentation
• Built an intuitive understanding of how it is achieved (learned) 
• Learned how it can be applied in a novel way for real life applications by building an action shot photo effects application 
• Gained appreciation and awareness for dealing with probability outcomes from machine learning models

Let's begin by better understanding what semantic segmentation is and get an intuitive understanding of how it is achieved.