Chapter 8: Fine-Grained Understanding of Images through Segmentation

Image segmentation is one of the biggest areas of study in computer vision. It consists of simplifying the visual contents of an image by grouping together pixels that share one or more defining characteristics, such as location, color, or texture. As is the case with many other subareas of computer vision, image segmentation has been greatly boosted by deep neural networks, mainly in industries such as medicine and autonomous driving.

While it's great to classify the contents of an image, more often than not, it's not enough. What if we want to know exactly where an object is? What if we're interested in its shape? What if we need its contour? These fine-grained needs cannot be met with traditional classification techniques. However, as we'll discover in this chapter, we can frame an image segmentation problem in a very similar way to a regular classification project. How? Instead of labeling...