cycle_gan_build_generator_model.py

def _encode__layer(self, input_layer, filters):
    layer = Conv2D(filters, kernel_size=4, strides=2, padding='same')(input_layer)
    layer = LeakyReLU(alpha=0.2)(layer)
    layer = InstanceNormalization()(layer)
    return layer

def _decode_transform_layer(self, input_layer, forward_layer, filters):
    layer = UpSampling2D(size=2)(input_layer)
    layer = Conv2D(filters, kernel_size=4, strides=1, padding='same', activation='relu')(layer)
    layer = InstanceNormalization()(layer)
    layer = Concatenate()([layer, forward_layer])
    return layer

def _build_generator_model(self):
    generator_input = Input(shape=self.img_shape)

    print("Build Encoder...")
    encode_layer_1 = self._encode__layer(generator_input, 32);
    encode_layer_2 = self._encode__layer(encode_layer_1, 64);
    encode_layer_3 = self._encode__layer(encode_layer_2, 128);
    encode_layer_4 = self._encode__layer(encode_layer_3, 256);

    print("Build Transformer - Decoder...")
    decode_transform_layer1 = self._decode_transform_layer(encode_layer_4, encode_layer_3, 128);
    decode_transform_layer2 = self._decode_transform_layer(decode_transform_layer1, encode_layer_2, 64);
    decode_transform_layer3 = self._decode_transform_layer(decode_transform_layer2, encode_layer_1, 32);

    generator_model = UpSampling2D(size = 2)(decode_transform_layer3)
    generator_model = Conv2D(self.img_shape[2], kernel_size=4, strides=1, padding='same', activation='tanh')(generator_model)

    final_generator_model = Model(generator_input, generator_model)
    final_generator_model.summary()
    return final_generator_model