GoogleNet Inception V3 model

As a concrete implementation of a CNN, in this section, I am going to present the GoogleNet (https://ai.google/research/pubs/pub43022) architecture by Google (https://www.google.com/) and its inception layers. It has been presented at the ImageNet Large Scale Visual Recognition Challenge 2014 (ILSVRC2014, http://www.image-net.org/challenges/LSVRC/2014/). Needless to say, it won that competition. The distinct characteristic of this implementation is the following: increased depth and width and, at the same time, a constant computational budget. Improved computing resources utilization is part of the network design.

This chart summarizes all of the layers for this network implementation presented in the context:

Figure 5.4: GoogleNet layers

There are 22 layers with parameters (excluding the pooling layers; the total is 27 if they are included) and almost...