def build_image_embeddings(self):
"""Builds the image model subgraph and generates image embeddings.
Inputs:
self.images
Outputs:
self.image_embeddings
"""
inception_output = image_embedding.inception_v3(
self.images,
trainable=self.train_inception,
is_training=self.is_training())
self.inception_variables = tf.get_collection(
tf.GraphKeys.GLOBAL_VARIABLES, scope="InceptionV3")
# Map inception output into embedding space.
with tf.variable_scope("image_embedding") as scope:
image_embeddings = tf.contrib.layers.fully_connected(
inputs=inception_output,
num_outputs=self.config.embedding_size,
activation_fn=None,
weights_initializer=self.initializer,
biases_initializer=None,
scope=scope)
# Save the embedding size in the graph.
tf.constant(self.config.embedding_size, name="embedding_size")
self.image_embeddings = image_embeddings
def build_image_embeddings(self):
"""Builds the image model subgraph and generates image embeddings.
Inputs:
self.images
Outputs:
self.image_embeddings
"""
inception_output = image_embedding.inception_v3(
self.images,
trainable=self.train_inception,
is_training=self.is_training())
self.inception_variables = tf.get_collection(
tf.GraphKeys.GLOBAL_VARIABLES, scope="InceptionV3")
# Map inception output into embedding space.
with tf.variable_scope("image_embedding") as scope:
image_embeddings = tf.contrib.layers.fully_connected(
inputs=inception_output,
num_outputs=self.config.embedding_size,
activation_fn=None,
weights_initializer=self.initializer,
biases_initializer=None,
scope=scope)
# Save the embedding size in the graph.
tf.constant(self.config.embedding_size, name="embedding_size")
self.image_embeddings = image_embeddings
def build_image_embeddings(self):
"""Builds the image model subgraph and generates image embeddings.
Inputs:
self.images
Outputs:
self.image_embeddings
"""
inception_output = image_embedding.inception_v3(
self.images,
trainable=self.train_inception,
is_training=self.is_training())
self.inception_variables = tf.get_collection(
tf.GraphKeys.GLOBAL_VARIABLES, scope="InceptionV3")
# Compute the average pool of the outputs from inception
context_tensor = tf.reduce_mean(inception_output, axis=[1, 2])
# Map inception output into embedding space.
with tf.variable_scope("image_embedding"):
image_embedding_map = tf.get_variable(
name="image_map",
shape=[context_tensor.shape[1], self.config.embedding_size],
initializer=self.initializer)
image_embeddings = tf.tensordot(context_tensor, image_embedding_map, 1)
# Save the embedding size in the graph.
tf.constant(self.config.embedding_size, name="embedding_size")
self.inception_output = inception_output
self.image_embedding_map = image_embedding_map
self.image_embeddings = image_embeddings
def testTrainableTrueIsTrainingFalse(self):
embeddings = image_embedding.inception_v3(
self._images, trainable=True, is_training=False)
self.assertEqual([self._batch_size, 2048], embeddings.get_shape().as_list())
self._verifyParameterCounts()
self._assertCollectionSize(376, tf.GraphKeys.GLOBAL_VARIABLES)
self._assertCollectionSize(188, tf.GraphKeys.TRAINABLE_VARIABLES)
self._assertCollectionSize(0, tf.GraphKeys.UPDATE_OPS)
self._assertCollectionSize(94, tf.GraphKeys.REGULARIZATION_LOSSES)
self._assertCollectionSize(0, tf.GraphKeys.LOSSES)
self._assertCollectionSize(23, tf.GraphKeys.SUMMARIES)
def testTrainableTrueIsTrainingFalse(self):
embeddings = image_embedding.inception_v3(
self._images, trainable=True, is_training=False)
self.assertEqual([self._batch_size, 2048], embeddings.get_shape().as_list())
self._verifyParameterCounts()
self._assertCollectionSize(376, tf.GraphKeys.GLOBAL_VARIABLES)
self._assertCollectionSize(188, tf.GraphKeys.TRAINABLE_VARIABLES)
self._assertCollectionSize(0, tf.GraphKeys.UPDATE_OPS)
self._assertCollectionSize(94, tf.GraphKeys.REGULARIZATION_LOSSES)
self._assertCollectionSize(0, tf.GraphKeys.LOSSES)
self._assertCollectionSize(23, tf.GraphKeys.SUMMARIES)
def build_image_embeddings(self):
"""Builds the image model subgraph and generates image embeddings.
Inputs:
self.images
Outputs:
self.image_embeddings
"""
if self.config.CNN_name == 'InceptionV3':
cnn_output = image_embedding.inception_v3(
self.images,
trainable=self.train_inception,
is_training=self.is_training())
elif self.config.CNN_name == 'InceptionV4':
cnn_output = image_embedding.inception_v4(
self.images,
trainable=self.train_inception,
is_training=self.is_training())
elif self.config.CNN_name == 'DenseNet':
cnn_output = image_embedding.DenseNet(x=self.images,
nb_blocks=None,
filters=24,
training=self.is_training(),
scope='DenseNet').model
elif self.config.CNN_name == 'ResNet':
raise ValueError(
"================== ResNet尚未实现！ ====================" %
self.config.CNN_name)
else:
raise ValueError(
'CNN_name [%s] was not recognized.(InceptionV3 InceptionV4 DenseNet ResNet)'
% self.config.CNN_name)
self.inception_variables = tf.get_collection(
tf.GraphKeys.GLOBAL_VARIABLES, scope=self.config.CNN_name)
# Map inception output into embedding space.
with tf.variable_scope("image_embedding") as scope:
image_embeddings = tf.contrib.layers.fully_connected(
inputs=cnn_output,
num_outputs=self.config.embedding_size,
activation_fn=None,
weights_initializer=self.initializer,
biases_initializer=None,
scope=scope)
# Save the embedding size in the graph.
tf.constant(self.config.embedding_size, name="embedding_size")
self.image_embeddings = image_embeddings
def make_image_embeddings(images):
inception_output = image_embedding.inception_v3(images,
trainable=False,
is_training=True)
initializer = tf.random_uniform_initializer(minval=-0.08, maxval=0.08)
# Map inception output into embedding space.
with tf.variable_scope("image_embedding") as scope:
image_embeddings = tf.contrib.layers.fully_connected(
inputs=inception_output,
num_outputs=512,
activation_fn=None,
weights_initializer=initializer,
biases_initializer=None,
scope=scope)
return inception_output, image_embeddings
def build_image_embeddings(self):
# Get image representation via Inception V3 model
inception_output_0 = image_embedding.inception_v3(
self.images_0,
trainable=self.train_inception,
is_training=self.is_training(),
scope="InceptionV3")
self.inception_variables = tf.get_collection(
tf.GraphKeys.GLOBAL_VARIABLES, scope="InceptionV3")
inception_output_1 = image_embedding.inception_v3(
self.images_1,
trainable=self.train_inception,
is_training=self.is_training(),
scope="InceptionV3")
self.inception_variables = tf.get_collection(
tf.GraphKeys.GLOBAL_VARIABLES, scope="InceptionV3")
inception_output_2 = image_embedding.inception_v3(
self.images_2,
trainable=self.train_inception,
is_training=self.is_training(),
scope="InceptionV3")
self.inception_variables = tf.get_collection(
tf.GraphKeys.GLOBAL_VARIABLES, scope="InceptionV3")
inception_output_3 = image_embedding.inception_v3(
self.images_3,
trainable=self.train_inception,
is_training=self.is_training(),
scope="InceptionV3")
self.inception_variables = tf.get_collection(
tf.GraphKeys.GLOBAL_VARIABLES, scope="InceptionV3")
inception_output_4 = image_embedding.inception_v3(
self.images_4,
trainable=self.train_inception,
is_training=self.is_training(),
scope="InceptionV3")
self.inception_variables = tf.get_collection(
tf.GraphKeys.GLOBAL_VARIABLES, scope="InceptionV3")
# Map inception output into embedding space.
with tf.variable_scope("image_embedding") as scope:
image_embeddings_0 = tf.contrib.layers.fully_connected(
inputs=inception_output_0,
num_outputs=self.config.embedding_size,
activation_fn=None,
weights_initializer=self.initializer,
biases_initializer=None,
scope=scope)
scope.reuse_variables()
image_embeddings_1 = tf.contrib.layers.fully_connected(
inputs=inception_output_1,
num_outputs=self.config.embedding_size,
activation_fn=None,
weights_initializer=self.initializer,
biases_initializer=None,
scope=scope)
image_embeddings_2 = tf.contrib.layers.fully_connected(
inputs=inception_output_2,
num_outputs=self.config.embedding_size,
activation_fn=None,
weights_initializer=self.initializer,
biases_initializer=None,
scope=scope)
image_embeddings_3 = tf.contrib.layers.fully_connected(
inputs=inception_output_3,
num_outputs=self.config.embedding_size,
activation_fn=None,
weights_initializer=self.initializer,
biases_initializer=None,
scope=scope)
image_embeddings_4 = tf.contrib.layers.fully_connected(
inputs=inception_output_4,
num_outputs=self.config.embedding_size,
activation_fn=None,
weights_initializer=self.initializer,
biases_initializer=None,
scope=scope)
tf.constant(self.config.embedding_size, name="embedding_size")
self.image_embeddings_0 = image_embeddings_0
self.image_embeddings_1 = image_embeddings_1
self.image_embeddings_2 = image_embeddings_2
self.image_embeddings_3 = image_embeddings_3
self.image_embeddings_4 = image_embeddings_4