def decoder(self, z, reuse=False):
with tf.variable_scope('decoder') as scope:
if reuse:
scope.reuse_variables()
z_dim = self.z_dim
if self.y_dim:
z = tf.concat(1, [z, self.y_labels])
z_dim = self.z_dim + self.y_dim
c_i = z
c_i = lin(c_i, 784, name='dec_lin_1')
# Now c_i has shape batch_size x 784
c_i = tf.maximum(0., c_i)
c_i = tf.reshape(c_i, [self.batch_size, 28, 28, 1])
# Now c_i has shape batch_size x 28 x 28 x 1
c_i = conv(c_i, 5, 1, 16, name="dec_conv_1")
c_i = tf.maximum(0., c_i)
# Now c_i has shape batch_size x 28 x 28 x 16
c_i = conv(c_i, 5, 1, 1, name="dec_conv_2")
# Now c_i has shape batch_size x 28 x 28 x 1
c_i = tf.reshape(c_i, shape=[self.batch_size, 784])
# Now c_i has shape batch_size x 784
y_image = tf.nn.sigmoid(c_i)
return y_image
def decoder(self, z, reuse=False):
with tf.variable_scope('decoder') as scope:
if reuse:
scope.reuse_variables()
z_dim = self.z_dim
if self.y_dim:
z = tf.concat(1, [z, self.y_labels])
z_dim = self.z_dim + self.y_dim
c_i = tf.reshape(z, shape=[self.batch_size, 1, 1, z_dim])
# Now c_i has shape batch_size x 1 x 1 x z_dim(+y_dim)
c_i = conv(c_i, 1, 1, 2048*self.k, name="dec_conv_1")
c_i = tf.reshape(c_i, shape=[self.batch_size, 4, 4, 128*self.k])
# Now c_i has shape batch_size x 4 x 4 x 128*k
c_i = lrelu_tconv(c_i, 4, 2, self.batch_size, 64*self.k, name="dec_tconv_2")
# Now c_i has shape batch_size x 8 x 8 x 64*k
c_i = bn_lrelu_tconv(c_i, 4, 2, self.batch_size, 12, self.bn_settings, name="dec_tconv_3")
# Now c_i has shape batch_size x 16 x 16 x 12
c_i = PS(c_i, 2, 3)
# Now c_i has shape batch_size x 32 x 32 x 3
y_image = tf.nn.tanh(c_i)
return y_image
def gan(self, x_image, reuse=False, features=False):
with tf.variable_scope('gan') as scope:
if reuse:
scope.reuse_variables()
c_i = x_image
# Now c_i has shape batch_size x 32 x 32 x self.channels
c_i = conv(c_i, 4, 2, 32 * self.k, name="gan_conv_1")
# Now c_i has shape batch_size x 16 x 16 x 128*k
c_i = lrelu_conv(c_i, 4, 2, 64 * self.k, name="gan_conv_2")
# c_i shape is batch_size x 8 x 8 x 256*k
c_i = bn_lrelu_conv(c_i, 4, 2, 128 * self.k, self.bn_settings, name="gan_conv_3")
# c_i shape is batch_size x 4 x 4 x 512*k
if features:
return c_i
c_i = bn_lrelu_conv(c_i, 4, 4, 1, self.bn_settings, name="gan_conv_fin")
# c_i shape is batch_size x 1 x 1 x 1
y = tf.reshape(c_i, shape=[self.batch_size, 1])
return y
def gan(self, x_image, reuse=False, features=False):
f = []
with tf.variable_scope('gan') as scope:
if reuse:
scope.reuse_variables()
c_i = x_image
c_i = tf.reshape(c_i, shape=[self.batch_size, 28, 28, 1])
# Now c_i has shape batch_size x 28 x 28 x 1
c_i = conv(c_i, 3, 1, 16 * self.k, name="enc_conv_1")
# Now c_i has shape batch_size x 28 x 28 x 16*k
c_i = bn_lrelu_conv(c_i, 4, 2, 32 * self.k, self.bn_settings, name="gan_conv_1")
# Now c_i has shape batch_size x 14 x 14 x 32*k
f.append(c_i)
if features:
return f
c_i = bn_lrelu_conv(c_i, 4, 2, 64 * self.k, self.bn_settings, name="gan_conv_2")
# Now c_i has shape batch_size x 7 x 7 x 64*k
c_i = bn_lrelu_conv(c_i, 7, 7, 1, self.bn_settings, name="gan_conv_fin")
# Now c_i has shape batch_size x 1 x 1 x z_dim
y = tf.reshape(c_i, shape=[self.batch_size, 1])
return y
def decoder(self, z, reuse=False):
with tf.variable_scope('decoder') as scope:
if reuse:
scope.reuse_variables()
z_dim = self.z_dim
if self.y_dim:
z = tf.concat(1, [z, self.y_labels])
z_dim = self.z_dim + self.y_dim
c_i = tf.reshape(z, shape=[self.batch_size, 1, 1, z_dim])
# Now c_i has shape batch_size x 1 x 1 x z_dim(+y_dim)
c_i = conv(c_i, 1, 1, 7*7*64*self.k, name="dec_conv_1")
c_i = tf.reshape(c_i, shape=[self.batch_size, 7, 7, 64*self.k])
# Now c_i has shape batch_size x 7 x 7 x 64*k
c_i = bn_lrelu_tconv(c_i, 4, 2, self.batch_size, 32 * self.k,
self.bn_settings, name="dec_tconv_2")
# Now c_i has shape batch_size x 14 x 14 x 32*k
c_i = bn_lrelu_tconv(c_i, 4, 2, self.batch_size, 16*self.k,
self.bn_settings, name="dec_tconv_3")
# Now c_i has shape batch_size x 28 x 28 x 16*k
c_i = bn_lrelu_conv(c_i, 3, 1, 1,
self.bn_settings, name="dec_conv_fin")
c_i = tf.reshape(c_i, shape=[self.batch_size, 784])
y_image = tf.nn.sigmoid(c_i)
return y_image
def decoder(self, z, reuse=False):
with tf.variable_scope('decoder') as scope:
if reuse:
scope.reuse_variables()
z_dim = self.z_dim
if self.y_dim:
z = tf.concat(1, [z, self.y_labels])
z_dim = self.z_dim + self.y_dim
c_i = tf.reshape(z, shape=[self.batch_size, 1, 1, z_dim])
# Now c_i has shape batch_size x 1 x 1 x z_dim(+y_dim)
c_i = conv(c_i,
filter_size=1,
stride=1,
out_channels=2048 * self.k,
name="dec_conv_1")
c_i = tf.reshape(c_i, shape=[self.batch_size, 4, 4, 128 * self.k])
# Now c_i has shape batch_size x 4 x 4 x 128*k
c_i = bn_lrelu_conv(c_i,
3,
1,
256 * self.k,
self.bn_settings,
name="dec_conv_2")
c_i = PS(c_i, 2, 64 * self.k)
# Now c_i has shape batch_size x 8 x 8 x 64*k
c_i = bn_lrelu_conv(c_i,
3,
1,
128 * self.k,
self.bn_settings,
name="dec_conv_3")
c_i = PS(c_i, 2, 32 * self.k)
# Now c_i has shape batch_size x 16 x 16 x 32*k
c_i = bn_lrelu_conv(c_i,
3,
1,
64,
self.bn_settings,
name="dec_conv_4")
c_i = PS(c_i, 2, 16)
# Now c_i has shape batch_size x 32 x 32 x 16*k
c_i = bn_lrelu_conv(c_i,
3,
1,
self.channels,
self.bn_settings,
name="dec_conv_fin")
y_image = tf.nn.tanh(c_i)
return y_image
def encoder(self):
c_i = self.x_image
c_i = tf.reshape(c_i, shape=[self.batch_size, 28, 28, 1])
# Now c_i has shape batch_size x 28 x 28 x 1
c_i = conv(c_i, 5, 1, 16, name="enc_conv_1")
c_i = tf.maximum(0., c_i)
# Now c_i has shape batch_size x 28 x 28 x 16
c_i = conv(c_i, 5, 1, 16, name="enc_conv_2")
c_i = tf.maximum(0., c_i)
# Now c_i has shape batch_size x 28 x 28 x 16
c_i = tf.reshape(c_i, shape=[self.batch_size, 16 * 784])
# Now c_i has shape batch_size x 784
c_i = lin(c_i, 400, name="enc_lin_1")
c_i = tf.maximum(0., c_i)
# Now c_i has shape batch_size x 400
z = lin(c_i, self.z_dim, name="enc_lin_2")
return z
def encoder(self):
c_i = self.x_image
# c_i shape is batch_size x 128 x 128 x self.channels
c_i = conv(c_i, 4, 2, 32 * self.k, name="enc_conv_1")
# c_i shape is batch_size x 64 x 64 x 32*k
c_i = bn_lrelu_conv(c_i,
4,
2,
64 * self.k,
self.bn_settings,
name="enc_conv_2")
# c_i shape is batch_size x 32 x 32 x 64*k
c_i = bn_lrelu_conv(c_i,
4,
2,
128 * self.k,
self.bn_settings,
name="enc_conv_3")
# c_i shape is batch_size x 16 x 16 x 128*k
c_i = bn_lrelu_conv(c_i,
4,
2,
256 * self.k,
self.bn_settings,
name="enc_conv_4")
# c_i shape is batch_size x 8 x 8 x 256*k
c_i = bn_lrelu_conv(c_i,
4,
2,
512 * self.k,
self.bn_settings,
name="enc_conv_5")
# c_i shape is batch_size x 4 x 4 x 512*k
c_i = bn_lrelu_conv(c_i,
4,
4,
self.z_dim,
self.bn_settings,
name="enc_conv_6")
# c_i shape is batch_size x 1 x 1 x z_dim
z = tf.reshape(c_i, shape=[self.batch_size, self.z_dim])
return z
def encoder(self):
c_i = self.x_image
c_i = tf.reshape(c_i, shape=[self.batch_size, 64, 64, 1])
# Now c_i has shape batch_size x 64 x 64 x self.channels
c_i = conv(c_i,
filter_size=3,
stride=1,
out_channels=16 * self.k,
name="enc_conv_1")
# Now c_i has shape batch_size x 64 x 64 x 16*k
c_i = lrelu_conv(c_i, 4, 2, 32 * self.k, name="enc_conv_2")
# Now c_i has shape batch_size x 32 x 32 x 32*k
c_i = bn_lrelu_conv(c_i,
4,
2,
64 * self.k,
self.bn_settings,
name="enc_conv_3")
# Now c_i has shape batch_size x 16 x 16 x 64*k
c_i = bn_lrelu_conv(c_i,
4,
2,
128 * self.k,
self.bn_settings,
name="enc_conv_4")
# Now c_i has shape batch_size x 8 x 8 x 128*k
c_i = bn_lrelu_conv(c_i,
4,
2,
256 * self.k,
self.bn_settings,
name="enc_conv_5")
# Now c_i has shape batch_size x 4 x 4 x 256*k
c_i = bn_lrelu_conv(c_i,
4,
4,
self.z_dim,
self.bn_settings,
name="enc_conv_fin")
# Now c_i has shape batch_size x 1 x 1 x z_dim
z = tf.reshape(c_i, shape=[self.batch_size, self.z_dim])
return z
def encoder(self):
c_i = self.x_image
# Now c_i has shape batch_size x 32 x 32 x self.channels
c_i = conv(c_i, 4, 2, 32*self.k, name="enc_conv_1")
# Now c_i has shape batch_size x 16 x 16 x 32*k
c_i = lrelu_conv(c_i, 4, 2, 64*self.k, name="enc_conv_2")
# Now c_i has shape batch_size x 8 x 8 x 64*k
c_i = bn_lrelu_conv(c_i, 4, 2, 128*self.k, self.bn_settings, name="enc_conv_3")
# Now c_i has shape batch_size x 4 x 4 x 128*k
c_i = bn_lrelu_conv(c_i, 4, 4, self.z_dim, self.bn_settings, name="enc_conv_fin")
# Now c_i has shape batch_size x 1 x 1 x z_dim
z = tf.reshape(c_i, shape=[self.batch_size, self.z_dim])
return z
def encoder(self):
c_i = self.x_image
c_i = tf.reshape(c_i, shape=[self.batch_size, 28, 28, 1])
# Now c_i has shape batch_size x 28 x 28 x 1
c_i = conv(c_i, 3, 1, 16*self.k, name="enc_conv_1")
# Now c_i has shape batch_size x 28 x 28 x 16*k
c_i = bn_lrelu_conv(c_i, 4, 2, 32*self.k, self.bn_settings, name="enc_conv_2")
# Now c_i has shape batch_size x 14 x 14 x 32*k
c_i = bn_lrelu_conv(c_i, 4, 2, 64*self.k, self.bn_settings, name="enc_conv_3")
# Now c_i has shape batch_size x 7 x 7 x 64*k
c_i = bn_lrelu_conv(c_i, 7, 7, self.z_dim, self.bn_settings, name="enc_conv_fin")
# Now c_i has shape batch_size x 1 x 1 x z_dim
z = tf.reshape(c_i, shape=[self.batch_size, self.z_dim])
return z
