Neural Network Projects with Python

Scalability

The second requirement of facial recognition systems is that it needs to be scalable. The model that we train must ultimately be able to scale to millions of different users, each with a unique face. Again, this is where CNNs fall short. Recall that in Chapter 4, Cats Versus Dogs – Image Classification Using CNNs, we trained a CNN to differentiate cats from dogs. This neural network is only able to identify and classify images of cats and dogs, and not of other animals, which it was not trained on. This means that if we were to use CNNs for facial recognition, we would have to train a separate neural network for each individual user. This would simply be unworkable from a scalability point of view! This would mean that Amazon ...