def testAtrousFullyConvolutionalValues(self):
"""Verify dense feature extraction with atrous convolution."""
nominal_stride = 32
for output_stride in [4, 8, 16, 32, None]:
with slim.arg_scope(resnet_utils.resnet_arg_scope()):
with tf.Graph().as_default():
with self.test_session() as sess:
tf.set_random_seed(0)
inputs = create_test_input(2, 81, 81, 3)
# Dense feature extraction followed by subsampling.
output, _ = self._resnet_small(inputs, None,
is_training=False,
global_pool=False,
output_stride=output_stride)
if output_stride is None:
factor = 1
else:
factor = nominal_stride // output_stride
output = resnet_utils.subsample(output, factor)
# Make the two networks use the same weights.
tf.get_variable_scope().reuse_variables()
# Feature extraction at the nominal network rate.
expected, _ = self._resnet_small(inputs, None,
is_training=False,
global_pool=False)
sess.run(tf.global_variables_initializer())
self.assertAllClose(output.eval(), expected.eval(),
atol=1e-4, rtol=1e-4)
def bottleneck(inputs, depth, depth_bottleneck, stride, rate=1,
outputs_collections=None, scope=None):
"""Bottleneck residual unit variant with BN before convolutions.
This is the full preactivation residual unit variant proposed in [2]. See
Fig. 1(b) of [2] for its definition. Note that we use here the bottleneck
variant which has an extra bottleneck layer.
When putting together two consecutive ResNet blocks that use this unit, one
should use stride = 2 in the last unit of the first block.
Args:
inputs: A tensor of size [batch, height, width, channels].
depth: The depth of the ResNet unit output.
depth_bottleneck: The depth of the bottleneck layers.
stride: The ResNet unit's stride. Determines the amount of downsampling of
the units output compared to its input.
rate: An integer, rate for atrous convolution.
outputs_collections: Collection to add the ResNet unit output.
scope: Optional variable_scope.
Returns:
The ResNet unit's output.
"""
with tf.variable_scope(scope, 'bottleneck_v2', [inputs]) as sc:
depth_in = slim.utils.last_dimension(inputs.get_shape(), min_rank=4)
preact = slim.batch_norm(inputs, activation_fn=tf.nn.relu, scope='preact')
if depth == depth_in:
shortcut = resnet_utils.subsample(inputs, stride, 'shortcut')
else:
shortcut = slim.conv2d(preact, depth, [1, 1], stride=stride,
normalizer_fn=None, activation_fn=None,
scope='shortcut')
residual = slim.conv2d(preact, depth_bottleneck, [1, 1], stride=1,
scope='conv1')
residual = resnet_utils.conv2d_same(residual, depth_bottleneck, 3, stride,
rate=rate, scope='conv2')
residual = slim.conv2d(residual, depth, [1, 1], stride=1,
normalizer_fn=None, activation_fn=None,
scope='conv3')
output = shortcut + residual
return slim.utils.collect_named_outputs(outputs_collections,
sc.original_name_scope,
output)
def _atrousValues(self, bottleneck):
"""Verify the values of dense feature extraction by atrous convolution.
Make sure that dense feature extraction by stack_blocks_dense() followed by
subsampling gives identical results to feature extraction at the nominal
network output stride using the simple self._stack_blocks_nondense() above.
Args:
bottleneck: The bottleneck function.
"""
blocks = [
resnet_utils.Block('block1', bottleneck, [(4, 1, 1), (4, 1, 2)]),
resnet_utils.Block('block2', bottleneck, [(8, 2, 1), (8, 2, 2)]),
resnet_utils.Block('block3', bottleneck, [(16, 4, 1), (16, 4, 2)]),
resnet_utils.Block('block4', bottleneck, [(32, 8, 1), (32, 8, 1)])
]
nominal_stride = 8
# Test both odd and even input dimensions.
height = 30
width = 31
with slim.arg_scope(resnet_utils.resnet_arg_scope()):
with slim.arg_scope([slim.batch_norm], is_training=False):
for output_stride in [1, 2, 4, 8, None]:
with tf.Graph().as_default():
with self.test_session() as sess:
tf.set_random_seed(0)
inputs = create_test_input(1, height, width, 3)
# Dense feature extraction followed by subsampling.
output = resnet_utils.stack_blocks_dense(inputs,
blocks,
output_stride)
if output_stride is None:
factor = 1
else:
factor = nominal_stride // output_stride
output = resnet_utils.subsample(output, factor)
# Make the two networks use the same weights.
tf.get_variable_scope().reuse_variables()
# Feature extraction at the nominal network rate.
expected = self._stack_blocks_nondense(inputs, blocks)
sess.run(tf.global_variables_initializer())
output, expected = sess.run([output, expected])
self.assertAllClose(output, expected, atol=1e-4, rtol=1e-4)
def testConv2DSameEven(self):
n, n2 = 4, 2
# Input image.
x = create_test_input(1, n, n, 1)
# Convolution kernel.
w = create_test_input(1, 3, 3, 1)
w = tf.reshape(w, [3, 3, 1, 1])
tf.get_variable('Conv/weights', initializer=w)
tf.get_variable('Conv/biases', initializer=tf.zeros([1]))
tf.get_variable_scope().reuse_variables()
y1 = slim.conv2d(x, 1, [3, 3], stride=1, scope='Conv')
y1_expected = tf.to_float([[14, 28, 43, 26],
[28, 48, 66, 37],
[43, 66, 84, 46],
[26, 37, 46, 22]])
y1_expected = tf.reshape(y1_expected, [1, n, n, 1])
y2 = resnet_utils.subsample(y1, 2)
y2_expected = tf.to_float([[14, 43],
[43, 84]])
y2_expected = tf.reshape(y2_expected, [1, n2, n2, 1])
y3 = resnet_utils.conv2d_same(x, 1, 3, stride=2, scope='Conv')
y3_expected = y2_expected
y4 = slim.conv2d(x, 1, [3, 3], stride=2, scope='Conv')
y4_expected = tf.to_float([[48, 37],
[37, 22]])
y4_expected = tf.reshape(y4_expected, [1, n2, n2, 1])
with self.test_session() as sess:
sess.run(tf.global_variables_initializer())
self.assertAllClose(y1.eval(), y1_expected.eval())
self.assertAllClose(y2.eval(), y2_expected.eval())
self.assertAllClose(y3.eval(), y3_expected.eval())
self.assertAllClose(y4.eval(), y4_expected.eval())
def testSubsampleFourByFour(self):
x = tf.reshape(tf.to_float(tf.range(16)), [1, 4, 4, 1])
x = resnet_utils.subsample(x, 2)
expected = tf.reshape(tf.constant([0, 2, 8, 10]), [1, 2, 2, 1])
with self.test_session():
self.assertAllClose(x.eval(), expected.eval())
def testSubsampleThreeByThree(self):
x = tf.reshape(tf.to_float(tf.range(9)), [1, 3, 3, 1])
x = resnet_utils.subsample(x, 2)
expected = tf.reshape(tf.constant([0, 2, 6, 8]), [1, 2, 2, 1])
with self.test_session():
self.assertAllClose(x.eval(), expected.eval())
def bottleneck(inputs,
depth,
depth_bottleneck,
stride,
rate=1,
outputs_collections=None,
scope=None,
use_bounded_activations=False):
"""Bottleneck residual unit variant with BN after convolutions.
This is the original residual unit proposed in [1]. See Fig. 1(a) of [2] for
its definition. Note that we use here the bottleneck variant which has an
extra bottleneck layer.
When putting together two consecutive ResNet blocks that use this unit, one
should use stride = 2 in the last unit of the first block.
Args:
inputs: A tensor of size [batch, height, width, channels].
depth: The depth of the ResNet unit output.
depth_bottleneck: The depth of the bottleneck layers.
stride: The ResNet unit's stride. Determines the amount of downsampling of
the units output compared to its input.
rate: An integer, rate for atrous convolution.
outputs_collections: Collection to add the ResNet unit output.
scope: Optional variable_scope.
use_bounded_activations: Whether or not to use bounded activations. Bounded
activations better lend themselves to quantized inference.
Returns:
The ResNet unit's output.
"""
with tf.variable_scope(scope, 'bottleneck_v1', [inputs]) as sc:
depth_in = slim.utils.last_dimension(inputs.get_shape(), min_rank=4)
if depth == depth_in:
shortcut = resnet_utils.subsample(inputs, stride, 'shortcut')
else:
shortcut = slim.conv2d(
inputs,
depth, [1, 1],
stride=stride,
activation_fn=tf.nn.relu6 if use_bounded_activations else None,
scope='shortcut')
residual = slim.conv2d(inputs, depth_bottleneck, [1, 1], stride=1,
scope='conv1')
residual = resnet_utils.conv2d_same(residual, depth_bottleneck, 3, stride,
rate=rate, scope='conv2')
residual = slim.conv2d(residual, depth, [1, 1], stride=1,
activation_fn=None, scope='conv3')
if use_bounded_activations:
# Use clip_by_value to simulate bandpass activation.
residual = tf.clip_by_value(residual, -6.0, 6.0)
output = tf.nn.relu6(shortcut + residual)
else:
output = tf.nn.relu(shortcut + residual)
return slim.utils.collect_named_outputs(outputs_collections,
sc.name,
output)
def testSubsampleFourByFour(self):
x = tf.reshape(tf.to_float(tf.range(16)), [1, 4, 4, 1])
x = resnet_utils.subsample(x, 2)
expected = tf.reshape(tf.constant([0, 2, 8, 10]), [1, 2, 2, 1])
with self.test_session():
self.assertAllClose(x.eval(), expected.eval())
def testStridingLastUnitVsSubsampleBlockEnd(self):
"""Compares subsampling at the block's last unit or block's end.
Makes sure that the final output is the same when we use a stride at the
last unit of a block vs. we subsample activations at the end of a block.
"""
block = resnet_v1.resnet_v1_block
blocks = [
block('block1', base_depth=1, num_units=2, stride=2),
block('block2', base_depth=2, num_units=2, stride=2),
block('block3', base_depth=4, num_units=2, stride=2),
block('block4', base_depth=8, num_units=2, stride=1),
]
# Test both odd and even input dimensions.
height = 30
width = 31
with slim.arg_scope(resnet_utils.resnet_arg_scope()):
with slim.arg_scope([slim.batch_norm], is_training=False):
for output_stride in [1, 2, 4, 8, None]:
with tf.Graph().as_default():
with self.test_session() as sess:
tf.set_random_seed(0)
inputs = create_test_input(1, height, width, 3)
# Subsampling at the last unit of the block.
output = resnet_utils.stack_blocks_dense(
inputs, blocks, output_stride,
store_non_strided_activations=False,
outputs_collections='output')
output_end_points = slim.utils.convert_collection_to_dict(
'output')
# Make the two networks use the same weights.
tf.get_variable_scope().reuse_variables()
# Subsample activations at the end of the blocks.
expected = resnet_utils.stack_blocks_dense(
inputs, blocks, output_stride,
store_non_strided_activations=True,
outputs_collections='expected')
expected_end_points = slim.utils.convert_collection_to_dict(
'expected')
sess.run(tf.global_variables_initializer())
# Make sure that the final output is the same.
output, expected = sess.run([output, expected])
self.assertAllClose(output, expected, atol=1e-4, rtol=1e-4)
# Make sure that intermediate block activations in
# output_end_points are subsampled versions of the corresponding
# ones in expected_end_points.
for i, block in enumerate(blocks[:-1:]):
output = output_end_points[block.scope]
expected = expected_end_points[block.scope]
atrous_activated = (output_stride is not None and
2 ** i >= output_stride)
if not atrous_activated:
expected = resnet_utils.subsample(expected, 2)
output, expected = sess.run([output, expected])
self.assertAllClose(output, expected, atol=1e-4, rtol=1e-4)
def testSubsampleThreeByThree(self):
x = tf.reshape(tf.to_float(tf.range(9)), [1, 3, 3, 1])
x = resnet_utils.subsample(x, 2)
expected = tf.reshape(tf.constant([0, 2, 6, 8]), [1, 2, 2, 1])
with self.test_session():
self.assertAllClose(x.eval(), expected.eval())
def testStridingLastUnitVsSubsampleBlockEnd(self):
"""Compares subsampling at the block's last unit or block's end.
Makes sure that the final output is the same when we use a stride at the
last unit of a block vs. we subsample activations at the end of a block.
"""
block = resnet_v1.resnet_v1_block
blocks = [
block('block1', base_depth=1, num_units=2, stride=2),
block('block2', base_depth=2, num_units=2, stride=2),
block('block3', base_depth=4, num_units=2, stride=2),
block('block4', base_depth=8, num_units=2, stride=1),
]
# Test both odd and even input dimensions.
height = 30
width = 31
with slim.arg_scope(resnet_utils.resnet_arg_scope()):
with slim.arg_scope([slim.batch_norm], is_training=False):
for output_stride in [1, 2, 4, 8, None]:
with tf.Graph().as_default():
with self.test_session() as sess:
tf.set_random_seed(0)
inputs = create_test_input(1, height, width, 3)
# Subsampling at the last unit of the block.
output = resnet_utils.stack_blocks_dense(
inputs,
blocks,
output_stride,
store_non_strided_activations=False,
outputs_collections='output')
output_end_points = slim.utils.convert_collection_to_dict(
'output')
# Make the two networks use the same weights.
tf.get_variable_scope().reuse_variables()
# Subsample activations at the end of the blocks.
expected = resnet_utils.stack_blocks_dense(
inputs,
blocks,
output_stride,
store_non_strided_activations=True,
outputs_collections='expected')
expected_end_points = slim.utils.convert_collection_to_dict(
'expected')
sess.run(tf.global_variables_initializer())
# Make sure that the final output is the same.
output, expected = sess.run([output, expected])
self.assertAllClose(output,
expected,
atol=1e-4,
rtol=1e-4)
# Make sure that intermediate block activations in
# output_end_points are subsampled versions of the corresponding
# ones in expected_end_points.
for i, block in enumerate(blocks[:-1:]):
output = output_end_points[block.scope]
expected = expected_end_points[block.scope]
atrous_activated = (output_stride is not None
and 2**i >= output_stride)
if not atrous_activated:
expected = resnet_utils.subsample(
expected, 2)
output, expected = sess.run([output, expected])
self.assertAllClose(output,
expected,
atol=1e-4,
rtol=1e-4)
def bottleneck(inputs,
depth,
depth_bottleneck,
stride,
rate=1,
outputs_collections=None,
scope=None,
use_bounded_activations=False,
attention_module=None):
"""Bottleneck residual unit variant with BN after convolutions.
This is the original residual unit proposed in [1]. See Fig. 1(a) of [2] for
its definition. Note that we use here the bottleneck variant which has an
extra bottleneck layer.
When putting together two consecutive ResNet blocks that use this unit, one
should use stride = 2 in the last unit of the first block.
Args:
inputs: A tensor of size [batch, height, width, channels].
depth: The depth of the ResNet unit output.
depth_bottleneck: The depth of the bottleneck layers.
stride: The ResNet unit's stride. Determines the amount of downsampling of
the units output compared to its input.
rate: An integer, rate for atrous convolution.
outputs_collections: Collection to add the ResNet unit output.
scope: Optional variable_scope.
use_bounded_activations: Whether or not to use bounded activations. Bounded
activations better lend themselves to quantized inference.
Returns:
The ResNet unit's output.
"""
with tf.compat.v1.variable_scope(scope, 'bottleneck_v1', [inputs]) as sc:
depth_in = slim.utils.last_dimension(inputs.get_shape(), min_rank=4)
if depth == depth_in:
shortcut = resnet_utils.subsample(inputs, stride, 'shortcut')
else:
shortcut = slim.conv2d(
inputs,
depth, [1, 1],
stride=stride,
activation_fn=tf.nn.relu6 if use_bounded_activations else None,
scope='shortcut')
residual = slim.conv2d(inputs,
depth_bottleneck, [1, 1],
stride=1,
scope='conv1')
residual = resnet_utils.conv2d_same(residual,
depth_bottleneck,
3,
stride,
rate=rate,
scope='conv2')
residual = slim.conv2d(residual,
depth, [1, 1],
stride=1,
activation_fn=None,
scope='conv3')
if use_bounded_activations:
# Use clip_by_value to simulate bandpass activation.
residual = tf.clip_by_value(residual, -6.0, 6.0)
# Add attention_module
if attention_module is not None:
residual = attach_attention_module(residual, attention_module,
scope)
output = tf.nn.relu6(shortcut + residual)
else:
# Add attention_module
if attention_module is not None:
residual = attach_attention_module(residual, attention_module,
scope)
output = tf.nn.relu(shortcut + residual)
return slim.utils.collect_named_outputs(outputs_collections, sc.name,
output)
def bottleneck(inputs,
depth,
depth_bottleneck,
stride,
rate=1,
outputs_collections=None,
scope=None):
"""Bottleneck residual unit variant with BN before convolutions.
This is the full preactivation residual unit variant proposed in [2]. See
Fig. 1(b) of [2] for its definition. Note that we use here the bottleneck
variant which has an extra bottleneck layer.
When putting together two consecutive ResNet blocks that use this unit, one
should use stride = 2 in the last unit of the first block.
Args:
inputs: A tensor of size [batch, height, width, channels].
depth: The depth of the ResNet unit output.
depth_bottleneck: The depth of the bottleneck layers.
stride: The ResNet unit's stride. Determines the amount of downsampling of
the units output compared to its input.
rate: An integer, rate for atrous convolution.
outputs_collections: Collection to add the ResNet unit output.
scope: Optional variable_scope.
Returns:
The ResNet unit's output.
"""
with tf.variable_scope(scope, 'bottleneck_v2', [inputs]) as sc:
depth_in = slim.utils.last_dimension(inputs.get_shape(), min_rank=4)
preact = slim.batch_norm(inputs,
activation_fn=tf.nn.relu,
scope='preact')
if depth == depth_in:
shortcut = resnet_utils.subsample(inputs, stride, 'shortcut')
else:
shortcut = slim.conv2d(preact,
depth, [1, 1],
stride=stride,
normalizer_fn=None,
activation_fn=None,
scope='shortcut')
residual = slim.conv2d(preact,
depth_bottleneck, [1, 1],
stride=1,
scope='conv1')
residual = resnet_utils.conv2d_same(residual,
depth_bottleneck,
3,
stride,
rate=rate,
scope='conv2')
residual = slim.conv2d(residual,
depth, [1, 1],
stride=1,
normalizer_fn=None,
activation_fn=None,
scope='conv3')
output = shortcut + residual
return slim.utils.collect_named_outputs(outputs_collections,
sc.original_name_scope, output)
def bottleneck(inputs,
depth,
depth_bottleneck,
stride,
rate=1,
outputs_collections=None,
scope=None):
"""Bottleneck residual unit variant with BN before convolutions.
This is the full preactivation residual unit variant proposed in [2]. See
Fig. 1(b) of [2] for its definition. Note that we use here the bottleneck
variant which has an extra bottleneck layer.
When putting together two consecutive ResNet blocks that use this unit, one
should use stride = 2 in the last unit of the first block.
Args:
inputs: A tensor of size [batch, height, width, channels].
depth: The depth of the ResNet unit output.
depth_bottleneck: The depth of the bottleneck layers.
stride: The ResNet unit's stride. Determines the amount of downsampling of
the units output compared to its input.
rate: An integer, rate for atrous convolution.
outputs_collections: Collection to add the ResNet unit output.
scope: Optional variable_scope.
Returns:
The ResNet unit's output.
"""
with tf.variable_scope(scope, 'bottleneck_v2', [inputs]) as sc:
depth_in = slim.utils.last_dimension(inputs.get_shape(), min_rank=4)
preact = slim.batch_norm(inputs,
activation_fn=tf.nn.relu,
scope='preact')
if depth == depth_in:
shortcut = resnet_utils.subsample(inputs, stride, 'shortcut')
else:
shortcut = slim.conv2d(preact,
depth, [1, 1],
stride=stride,
normalizer_fn=None,
activation_fn=None,
scope='shortcut')
def f1():
temp = slim.conv2d(preact,
depth_bottleneck, [1, 1],
stride=1,
scope='conv1')
return temp
def f2():
temp = slim.conv2d(preact,
depth_bottleneck, [1, 1],
stride=1,
normalizer_fn=None,
activation_fn=tf.nn.relu,
scope='conv1/pie')
codebook_conv_1 = tf.get_variable(name='conv1/pie/codebook',
shape=[
256,
])
temp = mapping_to_codebook(temp, codebook_conv_1)
return temp
flag = tf.get_variable(name='conv1/flag', initializer=True)
residual = tf.cond(flag, f1, f2, name="cond")
def h1():
temp = resnet_utils.conv2d_same(residual,
depth_bottleneck,
3,
stride,
rate=rate,
scope='conv2')
return temp
def h2():
if stride == 1:
temp = slim.conv2d(residual,
depth_bottleneck,
3,
stride=1,
rate=rate,
padding='SAME',
normalizer_fn=None,
activation_fn=tf.nn.relu,
scope='conv2/pie')
codebook_conv_2 = tf.get_variable(name='conv2/pie/codebook',
shape=[
256,
])
temp = mapping_to_codebook(temp, codebook_conv_2)
return temp
else:
kernel_size_effective = 3 + (3 - 1) * (rate - 1)
pad_total = kernel_size_effective - 1
pad_beg = pad_total // 2
pad_end = pad_total - pad_beg
inputs = tf.pad(
residual,
[[0, 0], [pad_beg, pad_end], [pad_beg, pad_end], [0, 0]])
temp = slim.conv2d(inputs,
depth_bottleneck,
3,
stride=stride,
rate=rate,
padding='VALID',
normalizer_fn=None,
activation_fn=tf.nn.relu,
scope='conv2/pie')
codebook_conv_2 = tf.get_variable(name='conv2/pie/codebook',
shape=[
256,
])
temp = mapping_to_codebook(temp, codebook_conv_2)
return temp
flag = tf.get_variable(name='conv2/flag', initializer=True)
residual = tf.cond(flag, h1, h2, name="cond2")
residual = slim.conv2d(residual,
depth, [1, 1],
stride=1,
normalizer_fn=None,
activation_fn=None,
scope='conv3')
output = shortcut + residual
return slim.utils.collect_named_outputs(outputs_collections, sc.name,
output)
