In this chapter, we saw how CNN models are built, including what loss functions to use. We looked at the CIFAR and Imagenet datasets, and saw how to train a CNN for the task of classifying the CIFAR10 dataset. In doing so, we were introduced to the TensorFlow data API, which makes the task of loading and transforming data easier. Finally, we looked at ways to help improve the quality of our trained model by talking about different methods of initialization and regularization.

In the next chapter, we will solve the more difficult tasks of object detection, semantics, and instance segmentation.