AutoEncoderBlocks.py

from tensorflow.keras import layers
import tensorflow_addons as tfa
from tensorflow import keras
import tensorflow as tf

def downsample(filters, size, alpha, apply_instancenorm=True):
    initializer = tf.random_normal_initializer(0., 0.02)

    result = keras.Sequential()
    result.add(layers.Conv2D(filters, size, strides=2, padding='same',
                             kernel_initializer=initializer, use_bias=False))

    if apply_instancenorm:
        result.add(tfa.layers.InstanceNormalization())

    result.add(layers.LeakyReLU(alpha))

    return result

def upsample(filters, size, apply_batch=False):
    initializer = tf.random_normal_initializer(0., 0.02)

    result = keras.Sequential()
    result.add(layers.Conv2DTranspose(filters, size, strides=2,
                                      padding='same',
                                      kernel_initializer=initializer,
                                      use_bias=False))

    if apply_batch:
        result.add(layers.BatchNormalization())#tfa.layers.InstanceNormalization())

    result.add(layers.ReLU())

    return result