def conv_block(self, xx, i, j):
"""
:param xx: input for the i-th block
:param i: i-th block
:param j: the j-th convolution block
:return: the ouput of the i-th conv block in the
"""
out = tf.add(
tf.nn.conv2d(
xx,
self.task_parameter["res" + str(i + 1) + "conv_w" +
str(j + 1)],
self.no_stride,
"SAME",
),
self.task_parameter["res" + str(i + 1) + "conv_bias" + str(j + 1)],
)
out = tf.contrib.layers.batch_norm(
out,
activation_fn=None,
variables_collections=self.var_collections,
scope="scope" + str(i + 1) + str(j + 1),
reuse=self.reuse,
)
return network_utils.leaky_relu(out, 0.1)
def conv_block(xx):
out = tcl.conv2d(
xx,
n_filters,
3,
activation_fn=None,
normalizer_fn=tcl.batch_norm,
variables_collections=self.var_coll,
)
return network_utils.leaky_relu(out, 0.1)
def conv_block(xx, i, j):
out = tf.nn.conv2d(
xx,
self.task_parameter["res" + str(i + 1) + "conv_w" +
str(j + 1)],
self.no_stride,
"SAME",
)
out = tf.contrib.layers.batch_norm(out, activation_fn=None)
return network_utils.leaky_relu(out, 0.1)
def conv_block(self, xx, i, j):
out = tf.add(
tf.nn.conv2d(
xx,
self.task_parameter["res" + str(i + 1) + "conv_w" +
str(j + 1)],
self.no_stride,
"SAME",
),
self.task_parameter["res" + str(i + 1) + "conv_bias" + str(j + 1)],
)
out = tf.contrib.layers.batch_norm(
out,
activation_fn=None,
variables_collections=self.var_collections,
scope="scope" + str(i + 1) + str(j + 1),
reuse=self.reuse,
)
return network_utils.leaky_relu(out, 0.1)