def network(self, seed, batch_size):
s1, s2, s4, s8, s16 = conv_sizes(self.output_size, layers=4, stride=2)
# 64, 32, 16, 8, 4 - for self.output_size = 64
# default architecture
# For Cramer: self.gf_dim = 64
z_ = linear(seed, self.dim * 8 * s16 * s16, self.prefix +
'h0_lin') # project random noise seed and reshape
h0 = tf.reshape(z_, [batch_size, s16, s16, self.dim * 8])
h0 = tf.nn.relu(self.g_bn0(h0))
h1 = deconv2d(h0, [batch_size, s8, s8, self.dim * 4],
name=self.prefix + 'h1')
h1 = tf.nn.relu(self.g_bn1(h1))
h2 = deconv2d(h1, [batch_size, s4, s4, self.dim * 2],
name=self.prefix + 'h2')
h2 = tf.nn.relu(self.g_bn2(h2))
h3 = deconv2d(h2, [batch_size, s2, s2, self.dim * 1],
name=self.prefix + 'h3')
h3 = tf.nn.relu(self.g_bn3(h3))
h4 = deconv2d(h3, [batch_size, s1, s1, self.c_dim],
name=self.prefix + 'h4')
return tf.nn.sigmoid(h4)
def network(self, seed, batch_size):
s1, s2, s4, s8, s16, s32 = conv_sizes(self.output_size,
layers=5,
stride=2)
# project `z` and reshape
z_ = linear(seed, self.dim * 16 * s32 * s32, self.prefix + 'h0_lin')
h0 = tf.reshape(z_, [-1, s32, s32, self.dim * 16])
h0 = tf.nn.relu(self.g_bn0(h0))
h1 = deconv2d(h0, [batch_size, s16, s16, self.dim * 8],
name=self.prefix + 'h1')
h1 = tf.nn.relu(self.g_bn1(h1))
h2 = deconv2d(h1, [batch_size, s8, s8, self.dim * 4],
name=self.prefix + 'h2')
h2 = tf.nn.relu(self.g_bn2(h2))
h3 = deconv2d(h2, [batch_size, s4, s4, self.dim * 2],
name=self.prefix + 'h3')
h3 = tf.nn.relu(self.g_bn3(h3))
h4 = deconv2d(h3, [batch_size, s2, s2, self.dim],
name=self.prefix + 'h4')
h4 = tf.nn.relu(self.g_bn4(h4))
h5 = deconv2d(h4, [batch_size, s1, s1, self.c_dim],
name=self.prefix + 'h5')
return tf.nn.sigmoid(h5)
def network(self, seed, batch_size):
from core.resnet import block, ops
s1, s2, s4, s8, s16, s32 = conv_sizes(self.output_size,
layers=5,
stride=2)
# project `z` and reshape
z_ = linear(seed, self.dim * 16 * s32 * s32, self.prefix + 'h0_lin')
h0 = tf.reshape(z_, [-1, self.dim * 16, s32, s32]) # NCHW format
h1 = block.ResidualBlock(self.prefix + 'res1',
16 * self.dim,
8 * self.dim,
3,
h0,
resample='up')
h2 = block.ResidualBlock(self.prefix + 'res2',
8 * self.dim,
4 * self.dim,
3,
h1,
resample='up')
h3 = block.ResidualBlock(self.prefix + 'res3',
4 * self.dim,
2 * self.dim,
3,
h2,
resample='up')
h4 = block.ResidualBlock(self.prefix + 'res4',
2 * self.dim,
self.dim,
3,
h3,
resample='up')
h4 = ops.batchnorm.Batchnorm('g_h4', [0, 2, 3], h4, fused=True)
h4 = tf.nn.relu(h4)
# h5 = lib.ops.conv2d.Conv2D('g_h5', dim, 3, 3, h4)
h5 = tf.transpose(h4, [0, 2, 3, 1]) # NCHW to NHWC
h5 = deconv2d(h5, [batch_size, s1, s1, self.c_dim], name='g_h5')
return tf.nn.sigmoid(h5)