def extract_features(self, preprocessed_inputs):
"""Extract features from preprocessed inputs.
Args:
preprocessed_inputs: a [batch, height, width, channels] float tensor
representing a batch of images.
Returns:
feature_maps: a list of tensors where the ith tensor has shape
[batch, height_i, width_i, depth_i]
Raises:
ValueError: depth multiplier is not supported.
"""
if self._depth_multiplier != 1.0:
raise ValueError('Depth multiplier not supported.')
preprocessed_inputs = shape_utils.check_min_image_dim(
129, preprocessed_inputs)
with tf.variable_scope(self._resnet_scope_name,
reuse=self._reuse_weights) as scope:
with slim.arg_scope(resnet_v1.resnet_arg_scope()):
with (slim.arg_scope(self._conv_hyperparams_fn())
if self._override_base_feature_extractor_hyperparams else
context_manager.IdentityContextManager()):
with slim.arg_scope([resnet_v1.bottleneck],
use_bounded_activations=self.
_use_bounded_activations):
_, activations = self._resnet_base_fn(
inputs=ops.pad_to_multiple(preprocessed_inputs,
self._pad_to_multiple),
num_classes=None,
is_training=None,
global_pool=False,
output_stride=None,
store_non_strided_activations=True,
scope=scope)
with slim.arg_scope(self._conv_hyperparams_fn()):
feature_maps = feature_map_generators.pooling_pyramid_feature_maps(
base_feature_map_depth=self._base_feature_map_depth,
num_layers=self._num_layers,
image_features={
'image_features':
self._filter_features(activations)['block3']
})
return feature_maps.values()
def extract_features(self, preprocessed_inputs):
"""Extract features from preprocessed inputs.
Args:
preprocessed_inputs: a [batch, height, width, channels] float tensor
representing a batch of images.
Returns:
feature_maps: a list of tensors where the ith tensor has shape
[batch, height_i, width_i, depth_i]
Raises:
ValueError: depth multiplier is not supported.
"""
if self._depth_multiplier != 1.0:
raise ValueError('Depth multiplier not supported.')
preprocessed_inputs = shape_utils.check_min_image_dim(
129, preprocessed_inputs)
with tf.variable_scope(
self._resnet_scope_name, reuse=self._reuse_weights) as scope:
with slim.arg_scope(resnet_v1.resnet_arg_scope()):
with (slim.arg_scope(self._conv_hyperparams_fn())
if self._override_base_feature_extractor_hyperparams else
context_manager.IdentityContextManager()):
with slim.arg_scope(
[resnet_v1.bottleneck],
use_bounded_activations=self._use_bounded_activations):
_, activations = self._resnet_base_fn(
inputs=ops.pad_to_multiple(preprocessed_inputs,
self._pad_to_multiple),
num_classes=None,
is_training=None,
global_pool=False,
output_stride=None,
store_non_strided_activations=True,
scope=scope)
with slim.arg_scope(self._conv_hyperparams_fn()):
feature_maps = feature_map_generators.pooling_pyramid_feature_maps(
base_feature_map_depth=self._base_feature_map_depth,
num_layers=self._num_layers,
image_features={
'image_features': self._filter_features(activations)['block3']
})
return feature_maps.values()
def test_get_expected_variable_names(self, replace_pool_with_conv):
image_features = {
'image_features': tf.random_uniform([4, 19, 19, 1024])
}
feature_maps = feature_map_generators.pooling_pyramid_feature_maps(
base_feature_map_depth=1024,
num_layers=6,
image_features=image_features,
replace_pool_with_conv=replace_pool_with_conv)
expected_pool_variables = set([
'Base_Conv2d_1x1_1024/weights',
'Base_Conv2d_1x1_1024/biases',
])
expected_conv_variables = set([
'Base_Conv2d_1x1_1024/weights',
'Base_Conv2d_1x1_1024/biases',
'Conv2d_0_3x3_s2_1024/weights',
'Conv2d_0_3x3_s2_1024/biases',
'Conv2d_1_3x3_s2_1024/weights',
'Conv2d_1_3x3_s2_1024/biases',
'Conv2d_2_3x3_s2_1024/weights',
'Conv2d_2_3x3_s2_1024/biases',
'Conv2d_3_3x3_s2_1024/weights',
'Conv2d_3_3x3_s2_1024/biases',
'Conv2d_4_3x3_s2_1024/weights',
'Conv2d_4_3x3_s2_1024/biases',
])
init_op = tf.global_variables_initializer()
with self.test_session() as sess:
sess.run(init_op)
sess.run(feature_maps)
actual_variable_set = set(
[var.op.name for var in tf.trainable_variables()])
if replace_pool_with_conv:
self.assertSetEqual(expected_conv_variables,
actual_variable_set)
else:
self.assertSetEqual(expected_pool_variables,
actual_variable_set)
def test_get_expected_feature_map_shapes(self, replace_pool_with_conv):
image_features = {
'image_features': tf.random_uniform([4, 19, 19, 1024])
}
feature_maps = feature_map_generators.pooling_pyramid_feature_maps(
base_feature_map_depth=1024,
num_layers=6,
image_features=image_features,
replace_pool_with_conv=replace_pool_with_conv)
expected_pool_feature_map_shapes = {
'Base_Conv2d_1x1_1024': (4, 19, 19, 1024),
'MaxPool2d_0_2x2': (4, 10, 10, 1024),
'MaxPool2d_1_2x2': (4, 5, 5, 1024),
'MaxPool2d_2_2x2': (4, 3, 3, 1024),
'MaxPool2d_3_2x2': (4, 2, 2, 1024),
'MaxPool2d_4_2x2': (4, 1, 1, 1024),
}
expected_conv_feature_map_shapes = {
'Base_Conv2d_1x1_1024': (4, 19, 19, 1024),
'Conv2d_0_3x3_s2_1024': (4, 10, 10, 1024),
'Conv2d_1_3x3_s2_1024': (4, 5, 5, 1024),
'Conv2d_2_3x3_s2_1024': (4, 3, 3, 1024),
'Conv2d_3_3x3_s2_1024': (4, 2, 2, 1024),
'Conv2d_4_3x3_s2_1024': (4, 1, 1, 1024),
}
init_op = tf.global_variables_initializer()
with self.test_session() as sess:
sess.run(init_op)
out_feature_maps = sess.run(feature_maps)
out_feature_map_shapes = {
key: value.shape
for key, value in out_feature_maps.items()
}
if replace_pool_with_conv:
self.assertDictEqual(expected_conv_feature_map_shapes,
out_feature_map_shapes)
else:
self.assertDictEqual(expected_pool_feature_map_shapes,
out_feature_map_shapes)
def test_get_expected_variable_names(self, replace_pool_with_conv):
image_features = {
'image_features': tf.random_uniform([4, 19, 19, 1024])
}
feature_maps = feature_map_generators.pooling_pyramid_feature_maps(
base_feature_map_depth=1024,
num_layers=6,
image_features=image_features,
replace_pool_with_conv=replace_pool_with_conv)
expected_pool_variables = set([
'Base_Conv2d_1x1_1024/weights',
'Base_Conv2d_1x1_1024/biases',
])
expected_conv_variables = set([
'Base_Conv2d_1x1_1024/weights',
'Base_Conv2d_1x1_1024/biases',
'Conv2d_0_3x3_s2_1024/weights',
'Conv2d_0_3x3_s2_1024/biases',
'Conv2d_1_3x3_s2_1024/weights',
'Conv2d_1_3x3_s2_1024/biases',
'Conv2d_2_3x3_s2_1024/weights',
'Conv2d_2_3x3_s2_1024/biases',
'Conv2d_3_3x3_s2_1024/weights',
'Conv2d_3_3x3_s2_1024/biases',
'Conv2d_4_3x3_s2_1024/weights',
'Conv2d_4_3x3_s2_1024/biases',
])
init_op = tf.global_variables_initializer()
with self.test_session() as sess:
sess.run(init_op)
sess.run(feature_maps)
actual_variable_set = set(
[var.op.name for var in tf.trainable_variables()])
if replace_pool_with_conv:
self.assertSetEqual(expected_conv_variables, actual_variable_set)
else:
self.assertSetEqual(expected_pool_variables, actual_variable_set)
def test_get_expected_feature_map_shapes(self, replace_pool_with_conv):
image_features = {
'image_features': tf.random_uniform([4, 19, 19, 1024])
}
feature_maps = feature_map_generators.pooling_pyramid_feature_maps(
base_feature_map_depth=1024,
num_layers=6,
image_features=image_features,
replace_pool_with_conv=replace_pool_with_conv)
expected_pool_feature_map_shapes = {
'Base_Conv2d_1x1_1024': (4, 19, 19, 1024),
'MaxPool2d_0_2x2': (4, 10, 10, 1024),
'MaxPool2d_1_2x2': (4, 5, 5, 1024),
'MaxPool2d_2_2x2': (4, 3, 3, 1024),
'MaxPool2d_3_2x2': (4, 2, 2, 1024),
'MaxPool2d_4_2x2': (4, 1, 1, 1024),
}
expected_conv_feature_map_shapes = {
'Base_Conv2d_1x1_1024': (4, 19, 19, 1024),
'Conv2d_0_3x3_s2_1024': (4, 10, 10, 1024),
'Conv2d_1_3x3_s2_1024': (4, 5, 5, 1024),
'Conv2d_2_3x3_s2_1024': (4, 3, 3, 1024),
'Conv2d_3_3x3_s2_1024': (4, 2, 2, 1024),
'Conv2d_4_3x3_s2_1024': (4, 1, 1, 1024),
}
init_op = tf.global_variables_initializer()
with self.test_session() as sess:
sess.run(init_op)
out_feature_maps = sess.run(feature_maps)
out_feature_map_shapes = {key: value.shape
for key, value in out_feature_maps.items()}
if replace_pool_with_conv:
self.assertDictEqual(expected_conv_feature_map_shapes,
out_feature_map_shapes)
else:
self.assertDictEqual(expected_pool_feature_map_shapes,
out_feature_map_shapes)
def extract_features(self, preprocessed_inputs):
"""Extract features from preprocessed inputs.
Args:
preprocessed_inputs: a [batch, height, width, channels] float tensor
representing a batch of images.
Returns:
feature_maps: a list of tensors where the ith tensor has shape
[batch, height_i, width_i, depth_i]
"""
preprocessed_inputs = shape_utils.check_min_image_dim(
33, preprocessed_inputs)
with tf.variable_scope('MobilenetV1',
reuse=self._reuse_weights) as scope:
with slim.arg_scope(
mobilenet_v1.mobilenet_v1_arg_scope(
is_training=None, regularize_depthwise=True)):
with (slim.arg_scope(self._conv_hyperparams_fn())
if self._override_base_feature_extractor_hyperparams else
context_manager.IdentityContextManager()):
_, image_features = mobilenet_v1.mobilenet_v1_base(
ops.pad_to_multiple(preprocessed_inputs,
self._pad_to_multiple),
final_endpoint='Conv2d_13_pointwise',
min_depth=self._min_depth,
depth_multiplier=self._depth_multiplier,
use_explicit_padding=self._use_explicit_padding,
scope=scope)
with slim.arg_scope(self._conv_hyperparams_fn()):
feature_maps = feature_map_generators.pooling_pyramid_feature_maps(
base_feature_map_depth=0,
num_layers=6,
image_features={
'image_features': image_features['Conv2d_11_pointwise']
})
return list(feature_maps.values())
