In this chapter, we will look at how we can modify the VGG network backbone to produce ResNet 34, a network from 2015. Looking back at our past few chapters, the difference between our 2D MNIST, CIFAR, and VGG networks is simply the number of blocks of 3x3 convolutions. Why stop at this point, though? Let’s make even larger networks! Next, we’re going to look at the ResNet family of networks, starting with ResNet 34.
Conceptually, we’re going to start with a similar base to the VGG network we were just looking at. If the backbone of our VGG network can be thought of as [2, ...
