def max_pool(x, name, kernel_size, strides, padding):
"""Max pooling layer on each GPU device.
Args:
x: A tensor of size [batch_size, height_in, width_in, channels].
name: The prefix of tensorflow variables defined in this layer.
kernel_size: A number indicating the size of pooling kernels.
strides: A number indicating the stride of the sliding window for
height and width.
padding: 'VALID' or 'SAME'.
Returns:
A tensor of size [batch_size, height_out, width_out, channels].
"""
return nn.max_pool(x, name, kernel_size, strides, padding)
def bottleneck(x,
name,
filters,
strides=None,
dilation=None,
is_training=True,
use_global_status=True):
"""Builds the bottleneck module in ResNet.
This function stack 3 convolutional layers and fuse the output with
the residual connection.
Args:
x: A tensor of size [batch_size, height_in, width_in, channels].
name: The prefix of tensorflow variables defined in this layer.
filters: A number indicating the number of output channels.
strides: A number indicating the stride of the sliding window for
height and width.
dilation: A number indicating the dilation factor for height and width.
is_training: If the tensorflow variables defined in this layer
would be used for training.
use_global_status: enable/disable use_global_status for batch
normalization. If True, moving mean and moving variance are updated
by exponential decay.
Returns:
A tensor of size [batch_size, height_out, width_out, channels_out].
"""
if strides is None and dilation is None:
raise ValueError('None of strides or dilation is specified, ' +
'set one of them to 1 or bigger number.')
elif strides > 1 and dilation is not None and dilation > 1:
raise ValueError('strides and dilation are both specified, ' +
'set one of them to 1 or None.')
with tf.variable_scope(name) as scope:
c_i = x.get_shape().as_list()[-1]
if c_i != filters * 4:
# Use a convolutional layer as residual connection when the
# number of input channels is different from output channels.
shortcut = nn.conv(x,
name='shortcut',
filters=filters * 4,
kernel_size=1,
strides=strides,
padding='VALID',
biased=False,
bn=True,
relu=False,
is_training=is_training,
use_global_status=use_global_status)
elif strides > 1:
# Use max-pooling as residual connection when the number of
# input channel is same as output channels, but stride is
# larger than 1.
shortcut = nn.max_pool(x,
name='shortcut',
kernel_size=1,
strides=strides,
padding='VALID')
else:
# Otherwise, keep the original input as residual connection.
shortcut = x
# Build the 1st convolutional layer.
x = nn.conv(x,
name='conv1',
filters=filters,
kernel_size=1,
strides=1,
padding='SAME',
biased=False,
bn=True,
relu=True,
is_training=is_training,
use_global_status=use_global_status)
if dilation is not None and dilation > 1:
# If dilation > 1, apply atrous conv to the 2nd convolutional layer.
x = nn.atrous_conv(x,
name='conv2',
filters=filters,
kernel_size=3,
dilation=dilation,
padding='SAME',
biased=False,
bn=True,
relu=True,
is_training=is_training,
use_global_status=use_global_status)
else:
padding = 'VALID' if strides > 1 else 'SAME'
x = nn.conv(x,
name='conv2',
filters=filters,
kernel_size=3,
strides=strides,
padding=padding,
biased=False,
bn=True,
relu=True,
is_training=is_training,
use_global_status=use_global_status)
# Build the 3rd convolutional layer (increase the channels).
x = nn.conv(x,
name='conv3',
filters=filters * 4,
kernel_size=1,
strides=1,
padding='SAME',
biased=False,
bn=True,
relu=False,
is_training=is_training,
use_global_status=use_global_status)
# Fuse the convolutional outputs with residual connection.
x = tf.add_n([x, shortcut], name='add')
x = tf.nn.relu(x, name='relu')
return x
def resnet_v1(x,
name,
filters=[64, 128, 256, 512],
num_blocks=[3, 4, 23, 3],
strides=[2, 1, 1, 1],
dilations=[None, None, 2, 2],
is_training=True,
use_global_status=True,
reuse=False):
"""Helper function to build ResNet.
Args:
x: A tensor of size [batch_size, height_in, width_in, channels].
name: The prefix of tensorflow variables defined in this network.
filters: A list of numbers indicating the number of output channels
(The output channels would be 4 times to the numbers).
strides: A list of numbers indicating the stride of the sliding window for
height and width.
dilation: A number indicating the dilation factor for height and width.
is_training: If the tensorflow variables defined in this layer
would be used for training.
use_global_status: enable/disable use_global_status for batch
normalization. If True, moving mean and moving variance are updated
by exponential decay.
reuse: enable/disable reuse for reusing tensorflow variables. It is
useful for sharing weight parameters across two identical networks.
Returns:
A tensor of size [batch_size, height_out, width_out, channels_out].
"""
if len(filters) != len(num_blocks) or len(filters) != len(strides):
raise ValueError('length of lists are not consistent')
with tf.variable_scope(name, reuse=reuse) as scope:
# Build conv1.
x = nn.conv(x,
name='conv1',
filters=64,
kernel_size=7,
strides=2,
padding='VALID',
biased=False,
bn=True,
relu=True,
is_training=is_training,
use_global_status=use_global_status)
bn = []
bn.append(x)
# Build pool1.
x = nn.max_pool(x,
name='pool1',
kernel_size=3,
strides=2,
padding='VALID')
# Build residual bottleneck blocks.
for ib in range(len(filters)):
for iu in range(num_blocks[ib]):
name_format = 'block{:d}/unit_{:d}/bottleneck_v1'
block_name = name_format.format(ib + 1, iu + 1)
c_o = filters[ib] # output channel
# Apply strides to the last block.
s = strides[ib] if iu == num_blocks[ib] - 1 else 1
d = dilations[ib]
if iu == num_blocks[ib] - 1:
bn.append(x)
x = bottleneck(x,
name=block_name,
filters=c_o,
strides=s,
dilation=d,
is_training=is_training,
use_global_status=use_global_status)
return x, bn
def _unet_builder(x,
mask,
name,
filters=[64, 128, 256, 512, 1024],
num_blocks=[2, 3, 3, 3, 3],
strides=[2, 2, 2, 2, 2],
is_training=True,
use_global_status=False,
reuse=False):
"""Helper function to construct UNet.
"""
if len(filters) != len(num_blocks)\
or len(filters) != len(strides):
raise ValueError('length of lists are not consistent')
with tf.variable_scope('Analyzer', reuse=reuse) as scope:
with tf.name_scope(name):
input_x = x
# Encoder.
shortcuts = []
not_ignore_masks = []
for ib in range(len(filters)):
for iu in range(num_blocks[ib]):
name_format = 'layer{:d}/unit_{:d}/encoder/'
block_name = name_format.format(ib + 1, iu + 1)
c_o = filters[ib] # output channel
# strides at the begginning
s = strides[ib] if iu == 0 else 1
padding = 'VALID' if s > 1 else 'SAME'
if ib == 0 and iu == 0:
x = []
for ix, in_x in enumerate(input_x):
x.append(
nn.conv(
in_x,
name=block_name + 'conv{:d}'.format(ix),
filters=int(c_o / 2),
#filters=c_o,
kernel_size=3,
strides=s,
padding=padding,
#biased=False,
#bn=True,
biased=True,
bn=False,
relu=False,
decay=0.99,
is_training=is_training,
use_global_status=use_global_status))
x = tf.concat(x, axis=-1, name=block_name + 'concat')
else:
x = nn.conv(
x,
name=block_name + 'conv',
filters=c_o,
kernel_size=3,
strides=s,
padding=padding,
#biased=False,
#bn=True,
biased=True,
bn=False,
relu=False,
decay=0.99,
is_training=is_training,
use_global_status=use_global_status)
if iu == 0:
mask = nn.max_pool(mask,
block_name + 'mask_pool',
3,
s,
padding=padding)
not_ignore_masks.append(1 - mask)
f = tf.multiply(x,
not_ignore_masks[-1],
name=block_name + 'masked_conv')
tf.add_to_collection('Analyzer/features', f)
x = tf.nn.relu(x)
print(x)
shortcuts.append(x)
# Decoder.
for ib in range(len(shortcuts) - 1, 0, -1):
for iu in range(num_blocks[ib - 1]):
n, h, w, c_o = shortcuts[ib - 1].get_shape().as_list()
name_format = 'layer{:d}/unit_{:d}/decoder/'
block_name = name_format.format(2 * len(filters) - ib,
iu + 1)
x = nn.conv(
x,
name=block_name + 'conv',
filters=c_o,
kernel_size=3,
strides=1,
padding='SAME',
#biased=False,
#bn=True,
biased=True,
bn=False,
relu=False,
decay=0.99,
is_training=is_training,
use_global_status=use_global_status)
f = tf.multiply(x,
not_ignore_masks[ib],
name=block_name + 'masked_conv')
tf.add_to_collection('Analyzer/features', f)
x = tf.nn.relu(x)
if iu == 0:
x = tf.image.resize_bilinear(x, [h, w])
x = tf.concat([x, shortcuts[ib - 1]], axis=-1)
print(x)
c_i = 0
for in_x in input_x:
c_i += in_x.get_shape().as_list()[-1]
x = nn.conv(x,
name='block5/fc',
filters=c_i,
kernel_size=1,
strides=1,
padding='SAME',
biased=True,
bn=False,
relu=False,
is_training=is_training)
x = tf.image.resize_bilinear(x, tf.shape(input_x[0])[1:3])
tf.add_to_collection('Analyzer/outputs', x)
return x