ops.py

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


class Sigmoid(layers.Layer):

    def __init__(self):
        super(Sigmoid, self).__init__()

    def call(self, inputs):
        return keras.activations.sigmoid(inputs)


class Tanh(layers.Layer):
    def __init__(self):
        super(Tanh, self).__init__()

    def call(self, inputs):
        return keras.activations.tanh(inputs)


class Conv2D(layers.Layer):
    def __init__(self, filters, kernel_size, strides=2):
        super(Conv2D, self).__init__()
        self.conv_op = layers.Conv2D(filters=filters,
                                     kernel_size=kernel_size,
                                     strides=strides,
                                     padding='same',
                                     kernel_initializer=keras.initializers.TruncatedNormal(stddev=0.02),
                                     use_bias=True,
                                     bias_initializer=keras.initializers.Constant(value=0.0))

    def call(self, inputs):
        return self.conv_op(inputs)


class BatchNorm(layers.Layer):
    def __init__(self, is_training=False):
        super(BatchNorm, self).__init__()
        self.bn = tf.keras.layers.BatchNormalization(epsilon=1e-5,
                                                     momentum=0.9,
                                                     scale=True,
                                                     trainable=is_training)

    def call(self, inputs, training):
        x = self.bn(inputs, training=training)
        return x


class DenseLayer(layers.Layer):
    def __init__(self, hidden_n, is_input=False):
        super(DenseLayer, self).__init__()

        self.fc_op = layers.Dense(hidden_n,
                                  kernel_initializer=keras.initializers.RandomNormal(stddev=0.02),
                                  bias_initializer=keras.initializers.Constant(value=0.0))

    def call(self, inputs):
        x = self.fc_op(inputs)

        return x


class UpConv2D(layers.Layer):
    def __init__(self, filters, kernel_size, strides):
        super(UpConv2D, self).__init__()
        self.up_conv_op = layers.Conv2DTranspose(filters,
                                                 kernel_size=kernel_size,
                                                 strides=strides,
                                                 padding='same',
                                                 kernel_initializer=keras.initializers.RandomNormal(stddev=0.02),
                                                 use_bias=True,
                                                 bias_initializer=keras.initializers.Constant(value=0.0))

    def call(self, inputs):
        x = self.up_conv_op(inputs)
        return x


def conv_cond_concat(x, y):
    """Concatenate conditioning vector on feature map axis."""
    x_shapes = tf.shape(x)
    y_shapes = tf.shape(y)
    y = tf.reshape(y, [-1, 1, 1, y_shapes[1]])
    y_shapes = tf.shape(y)
    return tf.concat([x, y*tf.ones([x_shapes[0], x_shapes[1], x_shapes[2], y_shapes[3]])], 3)