def discriminator_128(image, labels, df_dim, number_classes, act=tf.nn.relu):
"""Builds the discriminator graph.
Args:
image: The current batch of images to classify as fake or real.
labels: The corresponding labels for the images.
df_dim: The df dimension.
number_classes: The number of classes in the labels.
act: The activation function used in the discriminator.
Returns:
- A `Tensor` representing the logits of the discriminator.
- A list containing all trainable varaibles defined by the model.
"""
with tf.compat.v1.variable_scope(
'discriminator', reuse=tf.compat.v1.AUTO_REUSE) as dis_scope:
h0 = optimized_block(image, df_dim, 'd_optimized_block1',
act=act) # 64 * 64
h1 = block(h0, df_dim * 2, 'd_block2', act=act) # 32 * 32
h1 = ops.sn_non_local_block_sim(h1, name='d_ops') # 32 * 32
h2 = block(h1, df_dim * 4, 'd_block3', act=act) # 16 * 16
h3 = block(h2, df_dim * 8, 'd_block4', act=act) # 8 * 8
h4 = block(h3, df_dim * 16, 'd_block5', act=act) # 4 * 4
h5 = block(h4, df_dim * 16, 'd_block6', downsample=False, act=act)
h5_act = act(h5)
h6 = tf.reduce_sum(input_tensor=h5_act, axis=[1, 2])
output = ops.snlinear(h6, 1, name='d_sn_linear')
classification_output = ops.snlinear(h6,
flags.FLAGS.num_classes,
name='d_sn_linear_class')
if labels is None:
pseudo_labels = tf.argmax(classification_output, axis=1)
h_labels = ops.sn_embedding(pseudo_labels,
number_classes,
df_dim * 16,
name='d_embedding')
else:
h_labels = ops.sn_embedding(labels,
number_classes,
df_dim * 16,
name='d_embedding')
output += tf.reduce_sum(input_tensor=h6 * h_labels,
axis=1,
keepdims=True)
var_list = tf.compat.v1.get_collection(
tf.compat.v1.GraphKeys.TRAINABLE_VARIABLES, dis_scope.name)
return output, classification_output, var_list
def biggan_discriminator_128(image,
labels,
df_dim,
number_classes,
act=tf.nn.relu):
"""Builds the discriminator graph.
TODO(ilyak): debug, this implementation doesn't work as is.
...
Only position of the non local block changes
"""
with tf.compat.v1.variable_scope(
'discriminator', reuse=tf.compat.v1.AUTO_REUSE) as dis_scope:
h0 = optimized_block(image, df_dim, 'd_optimized_block1',
act=act) # 64 * 64
h0 = ops.sn_non_local_block_sim(h0, name='d_ops') # 64 * 64
h1 = block(h0, df_dim * 2, 'd_block2', act=act) # 32 * 32
h2 = block(h1, df_dim * 4, 'd_block3', act=act) # 16 * 16
h3 = block(h2, df_dim * 8, 'd_block4', act=act) # 8 * 8
h4 = block(h3, df_dim * 16, 'd_block5', act=act) # 4 * 4
h5 = block(h4, df_dim * 16, 'd_block6', downsample=False, act=act)
h5_act = act(h5)
h6 = tf.reduce_sum(input_tensor=h5_act, axis=[1, 2])
output = ops.snlinear(h6, 1, name='d_sn_linear')
classification_output = ops.snlinear(h6,
flags.FLAGS.num_classes,
name='d_sn_linear_class')
if labels is None:
pseudo_labels = tf.argmax(classification_output, axis=1)
h_labels = ops.sn_embedding(pseudo_labels,
number_classes,
df_dim * 16,
name='d_embedding')
else:
h_labels = ops.sn_embedding(labels,
number_classes,
df_dim * 16,
name='d_embedding')
output += tf.reduce_sum(input_tensor=h6 * h_labels,
axis=1,
keepdims=True)
var_list = tf.compat.v1.get_collection(
tf.compat.v1.GraphKeys.TRAINABLE_VARIABLES, dis_scope.name)
return output, classification_output, var_list
def test_sn_embedding_shapes(self):
"""Tests the spectrally normalized embedding layer.
When label = 10, embedding_size = 128, the
output shape should be [10, 128]
"""
if tf.executing_eagerly():
# `compute_spectral_norm` doesn't work when executing eagerly.
return
label = tf.ones([
10,
], dtype=tf.int32)
vector = ops.sn_embedding(label,
number_classes=1000,
embedding_size=128)
self.assertEqual([10, 128], vector.shape.as_list())