Hands-On Deep Learning Algorithms with Python

Defining the decoder

Define the decoder network for reconstructing the image. It consists of three fully connected networks, as follows:

with tf.name_scope('Decoder'):    #masked digit capsule    v_j = tf.reshape(masked_v, shape=(batch_size, -1))    #first fully connected layer     fc1 = tf.contrib.layers.fully_connected(v_j, num_outputs=512)    #second fully connected layer    fc2 = tf.contrib.layers.fully_connected(fc1, num_outputs=1024)    #reconstructed image    reconstructed_image = tf.contrib.layers.fully_connected(fc2, num_outputs=784, activation_fn=tf.sigmoid)