def discriminator(self, x, y=None, share_params=False, reuse=False, name=""):
with tf.variable_scope("discriminator-%s" % name, reuse=reuse):
if y is None:
x = tf.layers.flatten(x)
x = tf.concat([x, y], axis=1)
x = t.dense(x, self.height * self.width * self.channel,
name='disc-' + name + '-dense-0-y')
x = tf.reshape(x, self.image_shape)
else:
pass
# Using conv2d pooling instead of max_pool2d because of the speed.
# In the CoGAN paper, max_pool2d is used.
x = t.conv2d(x, f=self.df_dim, k=5, s=2, reuse=False, name='disc-' + name + '-conv2d-0')
x = t.prelu(x, reuse=False, name='disc-' + name + '-prelu-0')
# x = tf.nn.max_pool(x, ksize=2, strides=2, padding='SAME', name='disc' + name + '-max_pool2d-0')
x = t.conv2d(x, f=self.df_dim * 2, k=5, s=2, reuse=False, name='disc-' + name + '-conv2d-1')
x = t.batch_norm(x, is_train=False, name='disc-bn-0')
x = t.prelu(x, reuse=False, name='disc-' + name + '-prelu-1')
# x = tf.nn.max_pool(x, ksize=2, strides=2, padding='SAME', name='disc' + name + '-max_pool2d-1')
x = tf.layers.flatten(x)
x = t.dense(x, self.fc_unit, reuse=share_params, name='disc-dense-0')
x = t.batch_norm(x, is_train=share_params, name='disc-bn-1')
x = t.prelu(x, reuse=share_params, name='disc-prelu-2')
x = t.dense(x, 1, reuse=share_params, name='disc-dense-1')
return x
def generator(self, z, y=None, share_params=False, reuse=False, training=True, name=""):
if y is None:
x = tf.concat([z, y], axis=1)
else:
x = z
x = tf.layers.flatten(x)
x = tf.layers.dense(x, self.fc_unit, reuse=share_params, name='gen-dense-0')
x = t.prelu(x, reuse=share_params, name='gen-prelu-0')
x = tf.layers.dense(x, self.gf_dim * 8 * 7 * 7, reuse=share_params, name='gen-dense-1')
x = t.batch_norm(x, reuse=share_params, is_train=training, name='gen-bn-0')
x = t.prelu(x, reuse=share_params, name='gen-prelu-1')
x = tf.reshape(x, (self.batch_size, 7, 7, self.gf_dim * 8))
# x = deconv2d(x, f=self.gf_dim * 16, k=4, s=1, reuse=share_params, name='gen-deconv2d-0')
# x = batch_norm(x, reuse=share_params, training=training, name="gen-bn-0")
# x = prelu(x, reuse=share_params, name='gen-prelu-1')
x = t.deconv2d(x, f=self.gf_dim * 4, k=3, s=2, reuse=share_params, name='gen-deconv2d-1')
x = t.batch_norm(x, reuse=share_params, is_train=training, name="gen-bn-1")
x = t.prelu(x, reuse=share_params, name='gen-prelu-2')
x = t.deconv2d(x, f=self.gf_dim * 2, k=3, s=2, reuse=share_params, name='gen-deconv2d-2')
x = t.batch_norm(x, reuse=share_params, is_train=training, name="gen-bn-2")
x = t.prelu(x, reuse=share_params, name='gen-prelu-3')
with tf.variable_scope("generator-%s" % name, reuse=reuse):
x = t.deconv2d(x, f=self.channel, k=6, s=1, reuse=False, name='gen-' + name + '-deconv2d-3')
x = tf.nn.sigmoid(x, name='gen' + name + '-sigmoid-0')
return x
def generator(self, x, reuse=None, is_train=True):
"""
:param x: LR (Low Resolution) images, (-1, 96, 96, 3)
:param reuse: scope re-usability
:param is_train: is trainable, default True
:return: SR (Super Resolution) images, (-1, 384, 384, 3)
"""
with tf.variable_scope("generator", reuse=reuse):
def residual_block(x, f, name="", _is_train=True):
with tf.variable_scope(name):
shortcut = tf.identity(x, name='n64s1-shortcut')
x = t.conv2d(x, f, 3, 1, name="n64s1-1")
x = t.batch_norm(x, is_train=_is_train, name="n64s1-bn-1")
x = t.prelu(x, reuse=reuse, name='n64s1-prelu-1')
x = t.conv2d(x, f, 3, 1, name="n64s1-2")
x = t.batch_norm(x, is_train=_is_train, name="n64s1-bn-2")
x = tf.add(x, shortcut)
return x
x = t.conv2d(x, self.gf_dim, 9, 1, name='n64s1-1')
x = t.prelu(x, name='n64s1-prelu-1')
skip_conn = tf.identity(x, name='skip_connection')
# B residual blocks
for i in range(1, 17): # (1, 9)
x = residual_block(x,
self.gf_dim,
name='b-residual_block_%d' % i,
_is_train=is_train)
x = t.conv2d(x, self.gf_dim, 3, 1, name='n64s1-3')
x = t.batch_norm(x, is_train=is_train, name='n64s1-bn-3')
x = tf.add(x, skip_conn)
# sub-pixel conv2d blocks
for i in range(1, 3):
x = t.conv2d(x,
self.gf_dim * 4,
3,
1,
name='n256s1-%d' % (i + 2))
x = t.sub_pixel_conv2d(x, f=None, s=2)
x = t.prelu(x, name='n256s1-prelu-%d' % i)
x = t.conv2d(x, self.channel, 9, 1,
name='n3s1') # (-1, 384, 384, 3)
x = tf.nn.tanh(x)
return x
def generator(self, z, y=None, share_params=False, reuse=False, name=""):
x = t.dense(z, self.fc_g_unit, reuse=share_params, name='gen-fc-1')
x = t.batch_norm(x, reuse=share_params, name='gen-bn-1')
x = t.prelu(x, reuse=share_params, name='gen-prelu-1')
x = t.dense(x,
self.gf_dim * 8 * 7 * 7,
reuse=share_params,
name='gen-fc-2')
x = t.batch_norm(x, reuse=share_params, name='gen-bn-2')
x = t.prelu(x, reuse=share_params, name='gen-prelu-2')
x = tf.reshape(x, (-1, 7, 7, self.gf_dim * 8))
for i in range(1, 3):
x = t.deconv2d(x,
f=self.gf_dim * 4 // i,
k=3,
s=2,
reuse=share_params,
name='gen-deconv2d-%d' % i)
x = t.batch_norm(x, reuse=share_params, name="gen-bn-%d" % (i + 2))
x = t.prelu(x, reuse=share_params, name='gen-prelu-%d' % (i + 2))
"""
x = z # tf.concat([z, y], axis=1)
loop = 5
for i in range(1, loop):
x = t.dense(x, self.fc_g_unit, reuse=share_params, name='gen-fc-%d' % i)
x = t.batch_norm(x, reuse=share_params, name='gen-bn-%d' % i)
x = t.prelu(x, reuse=share_params, name='gen-prelu-%d' % i)
"""
with tf.variable_scope("generator-%s" % name, reuse=reuse):
x = t.deconv2d(x,
f=self.channel,
k=6,
s=1,
reuse=False,
name='gen-' + name + '-deconv2d-3')
x = tf.nn.sigmoid(x, name='gen' + name + '-sigmoid-1')
"""
x = t.dense(x, self.n_input, reuse=False, name='gen-' + name + '-fc-%d' % loop)
x = tf.nn.sigmoid(x)
"""
return x
def residual_block(x, f, name="", _is_train=True):
with tf.variable_scope(name):
shortcut = tf.identity(x, name='n64s1-shortcut')
x = t.conv2d(x, f, 3, 1, name="n64s1-1")
x = t.batch_norm(x, is_train=_is_train, name="n64s1-bn-1")
x = t.prelu(x, reuse=reuse, name='n64s1-prelu-1')
x = t.conv2d(x, f, 3, 1, name="n64s1-2")
x = t.batch_norm(x, is_train=_is_train, name="n64s1-bn-2")
x = tf.add(x, shortcut)
return x
def discriminator(self,
x,
y=None,
share_params=False,
reuse=False,
name=""):
with tf.variable_scope("discriminator-%s" % name, reuse=reuse):
x = tf.reshape(x, (-1, self.height, self.width, self.channel))
x = t.conv2d(x,
f=self.df_dim * 1,
k=5,
s=2,
reuse=False,
name='disc-' + name + '-conv2d-1')
x = t.prelu(x, reuse=False, name='disc-' + name + '-prelu-1')
# x = tf.layers.max_pooling2d(x, pool_size=2, strides=2, padding='SAME',
# name='disc-' + name + '-max_pool2d-1')
x = t.conv2d(x,
f=self.df_dim * 2,
k=5,
s=2,
reuse=False,
name='disc-' + name + '-conv2d-2')
x = t.batch_norm(x, reuse=reuse, name='disc-' + name + '-bn-1')
x = t.prelu(x, reuse=False, name='disc-' + name + '-prelu-2')
# x = tf.layers.max_pooling2d(x, pool_size=2, strides=2, padding='SAME',
# name='disc-' + name + '-max_pool2d-2')
x = tf.layers.flatten(x)
x = t.dense(x, self.fc_d_unit, reuse=share_params, name='disc-fc-1')
x = t.batch_norm(x, reuse=share_params, name='disc-bn-2')
x = t.prelu(x, reuse=share_params, name='disc-prelu-3')
x = t.dense(x, 1, reuse=share_params, name='disc-fc-2')
return x