With the help of TensorFlow and Keras, we caught up with years of research in deep learning. As CNNs have become central to modern computer vision (and machine learning in general), it is essential to understand how they perform, and what kinds of layers they are composed of. As presented in this chapter, TensorFlow and Keras provide clear interfaces to efficiently build such networks. They are also implementing several advanced optimization and regularization techniques (such as various optimizers, L1/L2 regularization, dropout, and batch normalization) to improve the performance and robustness of trained models, which is important to keep in mind for any application.
We now have the tools to finally tackle more challenging computer vision tasks.
In the next chapter, we will therefore present several CNN architectures applied to the task of classifying large picture datasets...
