Transforming images with encoders-decoders

As presented in Chapter 1, Computer Vision and Neural Networks, multiple typical tasks in computer vision require pixel-level results. For example, semantic segmentation methods classify each pixel of an image, and smart editing tools return images with some pixels altered (for example, to remove unwanted elements). In this section, we will present encoders-decoders, and how convolutional neural networks (CNNs) following this paradigm can be applied to such applications.