So far, we have discussed all the necessary building blocks for us to be able to implement solutions to common problems such as image classification and detection. In this chapter, we will talk about the implementation of some common model architectures that have shown high performance in many of these common tasks. These architectures have remained popular since they were first created, and they continue to be widely used today.
By the end of this chapter, you will gain an understanding of the different types of CNN models that exist, along with their use cases in a variety of different computer vision problems. While implementing these models, you will learn how these models were designed and the advantages for each of them. Finally, we will talk about how we can modify these architectures in order to make training and performance/efficiency better.
In summary, this chapter will cover the following topics:
- How to improve parameter efficiency
- How to implement the VGG nets in TensorFlow
- How to implement Inception nets in TensorFlow
- How to implement Residual nets in TensorFlow
- How to implement an architecture that is friendlier for mobile devices