Visualizing the output of the intermediate layers of a neural network

In the previous section, we built a model that learns to classify gender from images with an accuracy of 89%. However, as of now, it is a black box for us in terms of what the filters are learning.

In this section, we will learn how to extract what the various filters in a model are learning. Additionally, we will contrast the scenario of what the filters in the initial layers are learning with what the features in the last few layers are learning.

Getting ready

To understand how to extract what the various filters are learning, let's adopt the following strategy:

• We will select an image on which to perform analysis.
• We will select the first convolution...