We have just learned how to create neural networks that output predictions that are more complex than just a single class. In this chapter, we will push this concept further and introduce encoders-decoders, which are models used to edit or generate full images. We will present how encoder-decoder networks can be applied to a wide range of applications, from image denoising to object and instance segmentation. This chapter comes with several concrete examples, such as the application of encoders-decoders to semantic segmentation for self-driving cars.
The following topics will be covered in this chapter:
- What encoders-decoders are, and how they are trained for pixel-level prediction
- Which novel layers they ...
