Advanced Deep Learning with Python

Implementing Siamese networks

In this section, we'll use Keras to implement a simple example of Siamese networks, which will verify whether two MNIST images are from the same class or not. It is partially based on https://github.com/keras-team/keras/blob/master/examples/mnist_siamese.py.

Let's look at how to do this step by step:

1. We'll start with the import statements:

import randomimport numpy as npimport tensorflow as tf

2. Next, we'll implement the create_pairs function to create the train/test dataset (both for training and testing):

def create_pairs(inputs: np.ndarray, labels: np.ndarray):    num_classes = 10    digit_indices = [np.where(labels == i)[0] for i in range(num_classes)]    pairs = list()    labels = list()    n = min([len(digit_indices[d]) ...