LeNet with a normalization layer

lenet_with_normalization.py

# Create the input vector for  images
inputs = Input((WIDTH, HEIGHT))

# The first layer is the preprocessing layer, which is bound to the input vector
x = Normalize()(inputs)

# Implement LeNet
x = tf.layers.Conv2D(filters=6, kernel_size=(5,5), strides=1, activation='tanh', input_shape=(HEIGHT, HEIGHT, NUM_CHANNELS))(x)
x = tf.layers.AveragePooling2D(pool_size=(2,2))(x)
x = tf.layers.Conv2D(filters=16, kernel_size=(5,5), strides=1, activation='tanh')(x)
x = tf.layers.AveragePooling2D(pool_size=(2,2))(x)
x = tf.layers.Flatten()(x)
x = tf.layers.Dense(120)(x)
x = tf.layers.Dense(84)(x)

# Create an output layer for classifying the 10 digits
outputs = tf.layers.Dense(NCLASSES, activation='softmax')(x)

# Instantiate the model
model = Model(inputs, outputs)