5-dense_block.py

#!/usr/bin/env python3
"""Function that builds a dense block as described
in Densely Connected Convolutional Networks"""

import tensorflow.keras as K


def dense_block(X, nb_filters, growth_rate, layers):
    """Function that builds a dense block as described
    in Densely Connected Convolutional Networks"""
    concatenate = X
    for i in range(layers):
        X = K.layers.BatchNormalization(axis=3)(concatenate)
        X = K.layers.Activation('relu')(X)
        X = K.layers.Conv2D(growth_rate * 4, kernel_size=(1, 1),
                            padding='same',
                            strides=(1, 1),
                            kernel_initializer='he_normal')(X)
        X = K.layers.BatchNormalization(axis=3)(X)
        X = K.layers.Activation('relu')(X)
        X = K.layers.Conv2D(growth_rate, kernel_size=(3, 3), padding='same',
                            strides=(1, 1),
                            kernel_initializer='he_normal')(X)
        concatenate = K.layers.concatenate([concatenate, X], axis=3)
        nb_filters = nb_filters + growth_rate
    return (concatenate, nb_filters)