In the previous recipes, we implemented gender classification based on the VGG16 and VGG19 architectures. In this section, we'll implement the classification using the Inception architecture.
An intuition of how inception model comes in handy, is as follows.
There will be images where the object occupies the majority of the image. Similarly, there will be images where the object occupies a small portion of the total image. If we have the same size of kernels in both scenario, we are making it difficult for the model to learn – some images might have objects that are small and others might have objects that are larger.
To address this problem, we will have filters of multiple sizes that operate at the same layer.
In such a scenario, the network essentially gets wide rather than getting deep, as follows...
