def discriminator_dcgan(hparams, x, scope_name, train, reuse):
"""
The architecture is from
https://github.com/carpedm20/DCGAN-tensorflow
License: https://github.com/carpedm20/DCGAN-tensorflow/blob/master/LICENSE
"""
df_dim = 64 # dimension of discriminator filters in first conv layer
with tf.variable_scope(scope_name) as scope:
if reuse:
scope.reuse_variables()
d_bn1 = ops.batch_norm(name='d_bn1')
d_bn2 = ops.batch_norm(name='d_bn2')
d_bn3 = ops.batch_norm(name='d_bn3')
h0 = ops.lrelu(ops.conv2d(x, df_dim, name='d_h0_conv'))
h1 = ops.conv2d(h0, df_dim * 2, name='d_h1_conv')
h1 = ops.lrelu(d_bn1(h1, train=train))
h2 = ops.conv2d(h1, df_dim * 4, name='d_h2_conv')
h2 = ops.lrelu(d_bn2(h2, train=train))
h3 = ops.conv2d(h2, df_dim * 8, name='d_h3_conv')
h3 = ops.lrelu(d_bn3(h3, train=train))
h4 = ops.linear(tf.reshape(h3, [hparams.batch_size, -1]), 1,
'd_h3_lin')
d_logit = h4
d = tf.nn.sigmoid(d_logit)
return d, d_logit
def discriminator_fc(hparams, x, scope_name, train,
reuse): # pylint: disable = W0613
with tf.variable_scope(scope_name) as scope:
if reuse:
scope.reuse_variables()
d_bn0 = ops.batch_norm(name='d_bn0')
h0 = ops.linear(x, 100, 'd_h0_lin')
h0 = ops.lrelu(d_bn0(h0, train=train))
d_bn1 = ops.batch_norm(name='d_bn1')
h1 = ops.linear(h0, 100, 'd_h1_lin')
h1 = ops.lrelu(d_bn1(h1, train=train))
d_logit = ops.linear(h1, 1, 'd_h2_lin')
d = tf.nn.sigmoid(d_logit)
return d, d_logit
def generator_dcgan(hparams, y, z, scope_name, train, reuse):
"""
The architecture is from
https://github.com/carpedm20/DCGAN-tensorflow
License: https://github.com/carpedm20/DCGAN-tensorflow/blob/master/LICENSE
"""
s = 28
gf_dim = 64 # dimension of generator filters in first conv layer
gfc_dim = 1024 # dimension of generator units for for fully connected layer
with tf.variable_scope(scope_name) as scope:
if reuse:
scope.reuse_variables()
s2, s4 = int(s / 2), int(s / 4)
# yb = tf.expand_dims(tf.expand_dims(y, 1),2)
yb = tf.reshape(y, [hparams.batch_size, 1, 1, hparams.y_dim])
z = tf.concat([z, y], 1)
g_bn0 = ops.batch_norm(name='g_bn0')
h0 = ops.linear(z, gfc_dim, 'g_h0_lin')
h0 = tf.nn.relu(g_bn0(h0, train=train))
h0 = tf.concat([h0, y], 1)
g_bn1 = ops.batch_norm(name='g_bn1')
h1 = ops.linear(h0, gf_dim * 2 * s4 * s4, 'g_h1_lin')
h1 = tf.nn.relu(g_bn1(h1, train=train))
h1 = tf.reshape(h1, [hparams.batch_size, s4, s4, gf_dim * 2])
h1 = ops.conv_cond_concat(h1, yb)
g_bn2 = ops.batch_norm(name='g_bn2')
h2 = ops.deconv2d(h1, [hparams.batch_size, s2, s2, gf_dim * 2],
name='g_h2')
h2 = tf.nn.relu(g_bn2(h2, train=train))
h2 = ops.conv_cond_concat(h2, yb)
x_gen = tf.nn.sigmoid(
ops.deconv2d(h2, [hparams.batch_size, s, s, hparams.c_dim],
name='g_h3'))
return x_gen
def discriminator_dcgan(hparams, x, y, scope_name, train, reuse):
"""
The architecture is from
https://github.com/carpedm20/DCGAN-tensorflow
License: https://github.com/carpedm20/DCGAN-tensorflow/blob/master/LICENSE
"""
df_dim = 64 # dimension of discriminator filters in first conv layer
dfc_dim = 1024 # dimension of discriminator units for fully connected layer
with tf.variable_scope(scope_name) as scope:
if reuse:
scope.reuse_variables()
yb = tf.reshape(y, [hparams.batch_size, 1, 1, hparams.y_dim])
x = ops.conv_cond_concat(x, yb)
h0 = ops.lrelu(
ops.conv2d(x, hparams.c_dim + hparams.y_dim, name='d_h0_conv'))
h0 = ops.conv_cond_concat(h0, yb)
d_bn1 = ops.batch_norm(name='d_bn1')
h1 = ops.conv2d(h0, df_dim + hparams.y_dim, name='d_h1_conv')
h1 = ops.lrelu(d_bn1(h1, train=train))
h1 = tf.reshape(h1, [hparams.batch_size, -1])
h1 = tf.concat([h1, y], 1)
d_bn2 = ops.batch_norm(name='d_bn2')
h2 = ops.linear(h1, dfc_dim, 'd_h2_lin')
h2 = ops.lrelu(d_bn2(h2, train=train))
h2 = tf.concat([h2, y], 1)
h3 = ops.linear(h2, 1, 'd_h3_lin')
d_logit = h3
d = tf.nn.sigmoid(d_logit)
return d, d_logit
def generator_dcgan(hparams, z, scope_name, train, reuse):
"""
The architecture is from
https://github.com/carpedm20/DCGAN-tensorflow
License: https://github.com/carpedm20/DCGAN-tensorflow/blob/master/LICENSE
"""
gf_dim = 64 # dimension of generator filters in first conv layer
with tf.variable_scope(scope_name) as scope:
if reuse:
scope.reuse_variables()
s = 64
s2, s4, s8, s16 = int(s / 2), int(s / 4), int(s / 8), int(s / 16)
g_bn0 = ops.batch_norm(name='g_bn0')
g_bn1 = ops.batch_norm(name='g_bn1')
g_bn2 = ops.batch_norm(name='g_bn2')
g_bn3 = ops.batch_norm(name='g_bn3')
# project `z` and reshape
h0 = tf.reshape(ops.linear(z, gf_dim * 8 * s16 * s16, 'g_h0_lin'),
[-1, s16, s16, gf_dim * 8])
h0 = tf.nn.relu(g_bn0(h0, train=train))
h1 = ops.deconv2d(h0, [hparams.batch_size, s8, s8, gf_dim * 4],
name='g_h1')
h1 = tf.nn.relu(g_bn1(h1, train=train))
h2 = ops.deconv2d(h1, [hparams.batch_size, s4, s4, gf_dim * 2],
name='g_h2')
h2 = tf.nn.relu(g_bn2(h2, train=train))
h3 = ops.deconv2d(h2, [hparams.batch_size, s2, s2, gf_dim * 1],
name='g_h3')
h3 = tf.nn.relu(g_bn3(h3, train=train))
h4 = ops.deconv2d(h3, [hparams.batch_size, s, s, hparams.c_dim],
name='g_h4')
x_gen = tf.nn.tanh(h4)
return x_gen