def discriminator_forward(config,
incoming,
scope="discriminator",
name=None,
reuse=False):
with tf.variable_scope(scope, name, reuse=reuse):
output = conv_2d(incoming, config.gen_dim, 5, strides=2, name="conv1")
output = leaky_relu(output, 0.2)
output = conv_2d(output,
2 * config.gen_dim,
5,
strides=2,
name="conv2")
output = leaky_relu(output, 0.2)
output = conv_2d(output,
4 * config.gen_dim,
5,
strides=2,
name="conv3")
output = leaky_relu(output, 0.2)
output = conv_2d(output,
8 * config.gen_dim,
5,
strides=2,
name="conv4")
output = leaky_relu(output, 0.2)
output = tf.reshape(output, [-1, 4 * 4 * 8 * config.gen_dim])
output = fully_connected(output, 1, bias=False)
return output
def classifier_forward(config,
incoming,
name=None,
reuse=False,
scope="classifier"):
with tf.variable_scope(scope, name, reuse=reuse):
network = incoming
network = relu(
batch_normalization(
conv_2d(network,
32,
5,
activation='relu',
regularizer="L2",
strides=2)))
network = relu(
batch_normalization(
conv_2d(network,
64,
5,
activation='relu',
regularizer="L2",
strides=2)))
network = flatten(network)
network = relu(batch_normalization(fully_connected(network, 1024)))
network = dropout(network, 0.5)
network = fully_connected(network, 10)
return network
def discriminator_forward(config,
incoming,
scope="discriminator",
name=None,
reuse=False):
with tf.variable_scope(scope, name, reuse=reuse):
output = leaky_relu(conv_2d(incoming, config.dim, 5, 2), 0.2)
output = leaky_relu(conv_2d(output, 2 * config.dim, 5, 2), 0.2)
output = leaky_relu(conv_2d(output, 4 * config.dim, 5, 2), 0.2)
output = tf.reshape(output, [-1, 4 * 4 * 4 * config.dim])
output = tf.reshape(fully_connected(output, 1, bias=False), [-1])
return output
def generator_forward(config,
noise=None,
scope="generator",
name=None,
num_samples=-1,
reuse=False):
with tf.variable_scope(scope, name, reuse=reuse):
if noise is None:
noise = tf.random_normal(
[config.batch_size if num_samples == -1 else num_samples, 128],
name="noise")
output = fully_connected(noise, 4 * 4 * config.gen_dim, name="input")
output = tf.reshape(output, [-1, 4, 4, config.gen_dim])
output = residual_block_upsample(output, config.gen_dim, 3, name="rb1")
output = residual_block_upsample(output, config.gen_dim, 3, name="rb2")
output = residual_block_upsample(output, config.gen_dim, 3, name="rb3")
output = batch_normalization(output)
output = tf.nn.relu(output)
output = conv_2d(output, 3, 3, name="output")
output = tf.tanh(output)
return output
def discriminator_forward(config,
incoming,
scope="discriminator",
name=None,
reuse=False):
with tf.variable_scope(scope, name, reuse=reuse):
output = conv_2d(incoming, config.disc_dim, 3, name="input")
output = residual_block_downsample(output,
2 * config.disc_dim,
3,
name="rb1")
output = residual_block_downsample(output,
4 * config.disc_dim,
3,
name="rb2")
output = residual_block_downsample(output,
8 * config.disc_dim,
3,
name="rb3")
output = residual_block_downsample(output,
8 * config.disc_dim,
3,
name="rb4")
output = tf.reshape(output, [-1, 4 * 4 * 8 * config.disc_dim])
output = fully_connected(output, 1, name="output")
return tf.reshape(output, [-1])
def classifier_forward(config,
incoming,
name=None,
reuse=False,
scope="classifier"):
with tf.variable_scope(scope, name, reuse=reuse):
network = incoming
network = relu(conv_2d(network, 32, 5, strides=2))
network = relu(conv_2d(network, 64, 5, strides=2))
network = flatten(network)
network = relu(fully_connected(network, 1024))
network = dropout(network, 0.7)
network = fully_connected(network, 10)
return network
