def network(self, image, batch_size):
h0 = lrelu(conv2d(image, self.dim, name=self.prefix + 'h0_conv'))
h1 = lrelu(
self.d_bn1(conv2d(h0, self.dim * 2, name=self.prefix + 'h1_conv')))
h2 = lrelu(
self.d_bn2(conv2d(h1, self.dim * 4, name=self.prefix + 'h2_conv')))
h3 = lrelu(
self.d_bn3(conv2d(h2, self.dim * 8, name=self.prefix + 'h3_conv')))
hF = lrelu(
self.d_bn4(conv2d(h3, self.o_dim, name=self.prefix + 'hF_conv')))
hF = tf.reshape(hF, [batch_size, -1])
return {'h0': h0, 'h1': h1, 'h2': h2, 'h3': h3, 'hF': hF}
def network(self, image, batch_size):
o_dim = self.o_dim if (self.o_dim > 0) else 8 * self.dim
h0 = lrelu(conv2d(image, self.dim, name=self.prefix + 'h0_conv'))
h1 = lrelu(
self.d_bn1(conv2d(h0, self.dim * 2, name=self.prefix + 'h1_conv')))
h2 = lrelu(
self.d_bn2(conv2d(h1, self.dim * 4, name=self.prefix + 'h2_conv')))
h3 = lrelu(
self.d_bn3(conv2d(h2, self.dim * 8, name=self.prefix + 'h3_conv')))
hF = linear(tf.reshape(h3, [batch_size, -1]), o_dim,
self.prefix + 'h4_lin')
return {'h0': h0, 'h1': h1, 'h2': h2, 'h3': h3, 'hF': hF}
def network(self, image, batch_size):
from core.resnet import block, ops
image = tf.transpose(image, [0, 3, 1, 2]) # NHWC to NCHW
h0 = lrelu(
ops.conv2d.Conv2D(self.prefix + 'h0_conv',
3,
self.dim,
3,
image,
he_init=False))
h1 = block.ResidualBlock(self.prefix + 'res1',
self.dim,
2 * self.dim,
3,
h0,
resample='down')
h2 = block.ResidualBlock(self.prefix + 'res2',
2 * self.dim,
4 * self.dim,
3,
h1,
resample='down')
h3 = block.ResidualBlock(self.prefix + 'res3',
4 * self.dim,
8 * self.dim,
3,
h2,
resample='down')
h4 = block.ResidualBlock(self.prefix + 'res4',
8 * self.dim,
8 * self.dim,
3,
h3,
resample='down')
hF = tf.reshape(h4, [-1, 4 * 4 * 8 * self.dim])
hF = linear(hF, self.o_dim, self.prefix + 'h5_lin')
return {'h0': h0, 'h1': h1, 'h2': h2, 'h3': h3, 'h4': h4, 'hF': hF}