def generator(z_seed, is_training, init=False,reuse=False):
with tf.variable_scope('generator_model', reuse=reuse):
counter = {}
x = z_seed
with tf.variable_scope('dense_1'):
x = tf.layers.dense(x, units=4 * 4 * 512, kernel_initializer=init_kernel)
x = tf.layers.batch_normalization(x, training=is_training, name='batchnorm_1')
x = tf.nn.relu(x)
x = tf.reshape(x, [-1, 4, 4, 512])
with tf.variable_scope('deconv_1'):
x = tf.layers.conv2d_transpose(x, 256, [5, 5], strides=[2, 2], padding='SAME', kernel_initializer=init_kernel)
x = tf.layers.batch_normalization(x, training=is_training, name='batchnorm_2')
x = tf.nn.relu(x)
with tf.variable_scope('deconv_2'):
x = tf.layers.conv2d_transpose(x, 128, [5, 5], strides=[2, 2], padding='SAME', kernel_initializer=init_kernel)
x = tf.layers.batch_normalization(x, training=is_training, name='batchnormn_3')
x = tf.nn.relu(x)
with tf.variable_scope('deconv_3'):
output = nn.deconv2d(x, num_filters=3, filter_size=[5, 5], stride=[2, 2], nonlinearity=tf.tanh, init=init,
counters=counter, init_scale=0.1)
return output
def bad_sampler(self, z_seed, is_training=False, init=False):
with tf.variable_scope('bad_generator', reuse=tf.AUTO_REUSE):
counter = {}
x = z_seed
with tf.variable_scope('dense_1'):
# x = tf.layers.dense(x, units=4 * 4 * 512, kernel_initializer = init_kernel)
# x = tf.layers.batch_normalization(x, training=is_training, name='batchnorm_1')
x = nn._linear_fc(x, 4 * 4 * 512, 'bg_h0_lin')
x = nn.batch_norm_contrib(x,
name='batchnorm_1',
train=is_training)
x = tf.nn.relu(x)
x = tf.reshape(x, [-1, 4, 4, 512])
with tf.variable_scope('deconv_1'):
# x = tf.layers.conv2d_transpose(x, 256, [5, 5], strides=[2, 2], padding='SAME',
# kernel_initializer=init_kernel)
# x = tf.layers.batch_normalization(x, training=is_training, name='batchnorm_2')
x = nn._deconv2d(x,
256,
k_w=5,
k_h=5,
d_w=2,
d_h=2,
name='bg_dconv0') # [8, 8]
x = nn.batch_norm_contrib(x,
name='batchnorm_2',
train=is_training)
x = tf.nn.relu(x)
with tf.variable_scope('deconv_2'):
# x = tf.layers.conv2d_transpose(x, 128, [5, 5], strides=[2, 2], padding='SAME',
# kernel_initializer=init_kernel)
# x = tf.layers.batch_normalization(x, training=is_training, name='batchnormn_3')
x = nn._deconv2d(x,
128,
k_w=5,
k_h=5,
d_w=2,
d_h=2,
name='bg_dconv1')
x = nn.batch_norm_contrib(x,
name='batchnorm_2',
train=is_training)
x = tf.nn.relu(x)
with tf.variable_scope('deconv_3'):
output = nn.deconv2d(x,
num_filters=3,
filter_size=[5, 5],
stride=[2, 2],
nonlinearity=tf.tanh,
init=init,
counters=counter,
init_scale=0.1)
return output
def _build_generator(self,
tensor=None,
label=None,
training=False,
batch_norm=None,
init=False):
assert self.n_pixel % 16 == 0, "isize has to be a multiple of 16"
nfilt = 512
csize = 4
if label is None:
if self.input_y_g is None:
# TO BE EDITED
# sample generated images
batch_size = (self.input_z_g).shape[0]
#label = get_one_hot(np.repeat(np.tile(self.n_class), (batch_size // self.n_class) + 1), depth=self.n_class)
#label = label[0:batch_size, :]
label = get_one_hot(
np.random.randint(self.n_class, size=batch_size),
self.n_class)
else:
# get label from input
label = self.input_y_g
if tensor is None:
# add label to noise
tensor = tf.concat([self.input_z_g, label], 1)
else:
# assuming tensor is a specific noise
tensor = tf.concat([tensor, label], 1)
# tensor = tf.concat([tensor, tf.one_hot(label, self.n_class)], 1)
if batch_norm is None:
batch_norm = self.batch_norm_G
if batch_norm:
def bn(x, name=None):
return tf.contrib.layers.batch_norm(x,
is_training=training,
renorm=BATCH_RENORM,
decay=BATCH_NORM_DECAY)
else:
# return the same if bn is not aactivated
bn = tf.identity
with tf.variable_scope('generator', reuse=tf.AUTO_REUSE) as scope:
# set reuse if necessary
#if tf.get_collection(tf.GraphKeys.GLOBAL_VARIABLES, scope=scope.name):
# scope.reuse_variables()
#if constructed first time, it makes nothing reuse. only reusable on second call
# initial layer
with tf.variable_scope('initial.{0}-{1}'.format(
self.n_noise + self.n_class, nfilt)):
tensor = tf.reshape(tf.nn.relu(
bn(tf.layers.dense(tf.reshape(
tensor, [-1, 1, 1, self.n_noise + self.n_class]),
units=4 * 4 * 512,
kernel_initializer=init_normal(0.05),
name='dense'),
name='bn')),
shape=[-1, 4, 4, 512])
# upscaling layers
while csize < self.n_pixel / 2:
with tf.variable_scope('pyramid.{0}-{1}'.format(
nfilt, nfilt // 2)):
tensor = convconcatlayer(tensor, label)
tensor = tf.nn.relu(
bn(tf.layers.conv2d_transpose(
tensor,
nfilt // 2,
5,
2,
'same',
use_bias=not batch_norm,
kernel_initializer=init_normal(0.05),
name='conv'),
name='bn'))
csize *= 2
nfilt //= 2
# final layer
with tf.variable_scope('final.{0}-{1}'.format(
nfilt, self.n_channel)):
tensor = weightnorm.deconv2d(tensor,
self.n_channel, [5, 5], [2, 2],
'SAME',
init=init,
nonlinearity=tf.tanh)
# removed mask layer
return tensor, label
