def build_inputs(self):
input_queue = inputs.prefetch_input_data(
self.reader,
self.config.input_file_pattern,
batch_size=self.config.batch_size,
values_per_shard=self.config.values_per_input_shard,
input_queue_capacity_factor=self.config.input_queue_capacity_factor,
num_reader_threads=self.config.num_input_reader_threads)
assert self.config.num_preprocess_threads % 2 == 0
data = []
for thread_id in range(self.config.num_preprocess_threads):
# thread
serialized_example = input_queue.dequeue()
# caption id
sample, label, sentences_length = inputs.parse_example(
serialized_example,
sample_feature=self.config.sample_feature,
label_feature=self.config.label_feature,
sentences_length_feature=self.config.sentences_length_feature)
#
sample = data_preprocess(sample, sentences_length)
data.append([sample, label, sentences_length])
# mutil threads preprocessing the image
queue_capacity = (2 * self.config.num_preprocess_threads *
self.config.batch_size)
# pipe for batch data
#size: samples:[batch_size,sentence_num,sentence_len], labels:[batch_size],sentences_length:[batch_size,sentences_num]
samples, labels, sentences_length = inputs.batch_with_dynamic_pad(
data, batch_size=self.config.batch_size, queue_capacity=queue_capacity)
def build_inputs(self):
"""Input prefetching, preprocessing and batching.
Outputs:
self.images
self.input_seqs
self.target_seqs (training and eval only)
self.input_mask (training and eval only)
"""
if self.mode == "inference":
# In inference mode, images and inputs are fed via placeholders.
image_feed = tf.placeholder(dtype=tf.string, shape=[], name="image_feed")
input_feed = tf.placeholder(dtype=tf.int64,
shape=[None], # batch_size
name="input_feed")
# Process image and insert batch dimensions.
images = tf.expand_dims(self.process_image(image_feed), 0)
input_seqs = tf.expand_dims(input_feed, 1)
# No target sequences or input mask in inference mode.
target_seqs = None
input_mask = None
else:
# Prefetch serialized SequenceExample protos.
input_queue = input_ops.prefetch_input_data(
self.reader,
self.config.input_file_pattern,
is_training=self.is_training(),
batch_size=self.config.batch_size,
values_per_shard=self.config.values_per_input_shard,
input_queue_capacity_factor=self.config.input_queue_capacity_factor,
num_reader_threads=self.config.num_input_reader_threads)
# Image processing and random distortion. Split across multiple threads
# with each thread applying a slightly different distortion.
assert self.config.num_preprocess_threads % 2 == 0
images_and_captions = []
for thread_id in range(self.config.num_preprocess_threads):
serialized_sequence_example = input_queue.dequeue()
encoded_image, caption = input_ops.parse_sequence_example(
serialized_sequence_example,
image_feature=self.config.image_feature_name,
caption_feature=self.config.caption_feature_name)
image = self.process_image(encoded_image, thread_id=thread_id)
images_and_captions.append([image, caption])
# Batch inputs.
queue_capacity = (2 * self.config.num_preprocess_threads *
self.config.batch_size)
images, input_seqs, target_seqs, input_mask = (
input_ops.batch_with_dynamic_pad(images_and_captions,
batch_size=self.config.batch_size,
queue_capacity=queue_capacity))
self.images = images
self.input_seqs = input_seqs
self.target_seqs = target_seqs
self.input_mask = input_mask
def read_inputs(self):
if self.mode == 'inference':
image_feed = tf.placeholder(dtype=tf.string,
shape=[],
name="image_feed")
input_feed = tf.placeholder(dtype=tf.int64,
shape=[None],
name="input_feed")
images = tf.expand_dims(
image_processing.process_image(image_feed,
is_training=False,
height=299,
width=299,
thread_id=0,
image_format='jpeg'), 0)
input_seqs = tf.expand_dims(input_feed, 1)
target_seqs = None
input_mask = None
else:
input_queue = inputs_ops.prefetch_input_data(
self.reader,
file_pattern=self.input_file_pattern, # Must be given as input
is_training=True,
batch_size=32,
values_per_shard=2300,
input_queue_capacity_factor=2,
num_reader_threads=1)
images_and_captions = []
for thread_id in range(4):
serialized_sequence_example = input_queue.dequeue()
encoded_image, caption = inputs_ops.parse_sequence_example(
serialized_sequence_example,
image_feature="image/data",
caption_feature="image/caption_ids")
image = image_processing.process_image(encoded_image,
is_training=False,
height=299,
width=299,
thread_id=thread_id,
image_format="jpeg")
images_and_captions.append([image, caption])
queue_capacity = (2 * 4 * self.batch_size)
images, input_seqs, target_seqs, input_mask = (
inputs_ops.batch_with_dynamic_pad(
images_and_captions,
batch_size=32,
queue_capacity=queue_capacity))
self.images = images
self.input_seqs = input_seqs
self.target_seqs = target_seqs
self.input_mask = input_mask
def build_inputs(self):
input_queue = inputs.prefetch_input_data(
self.reader,
self.config.input_file_pattern,
batch_size=self.config.batch_size,
values_per_shard=self.config.values_per_input_shard,
input_queue_capacity_factor=self.config.
input_queue_capacity_factor,
num_reader_threads=self.config.num_input_reader_threads,
is_training=self.is_training())
assert self.config.num_preprocess_threads % 2 == 0
data = []
for thread_id in range(self.config.num_preprocess_threads):
# thread
serialized_example = input_queue.dequeue()
# caption id
sample, label, sentences_length = inputs.parse_example(
serialized_example,
sample_feature=self.config.sample_feature,
label_feature=self.config.label_feature,
sentences_length_feature=self.config.sentences_length_feature)
def data_preprocess(sample, sentences_length):
# sample,sentences_length
partition_num = sentences_length.shape[0]
partition = []
for i, sentence_length in enumerate(sentences_length):
partition.extend(np.ones(sentence_length) * i)
return tf.dynamic_partition(sample, partition, partition_num)
sample = data_preprocess(sample, sentences_length)
data.append([sample, label, sentences_length])
# mutil threads preprocessing the image
queue_capacity = (2 * self.config.num_preprocess_threads *
self.config.batch_size)
# pipe for batch data
# size: samples:[batch_size,sentence_num,sentence_len], labels:[batch_size],sentences_length:[batch_size,sentences_num]
self.samples, self.labels, self.sentences_length = inputs.batch_with_dynamic_pad(
data,
batch_size=self.config.batch_size,
queue_capacity=queue_capacity)
def build_input(self):
if self.config.mode == "inference":
input_seqs = Input(shape=(1,), dtype='int32')
images = Input(shape=(self.config.image_height, self.config.image_width,3,), dtype='float32')
else:
# Prefetch serialized SequenceExample protos.
input_queue = input_ops.prefetch_input_data(
self.reader,
self.config.input_file_pattern,
is_training=self.config.is_training(),
batch_size=self.config.batch_size,
values_per_shard=self.config.values_per_input_shard,
input_queue_capacity_factor=self.config.input_queue_capacity_factor,
num_reader_threads=self.config.num_input_reader_threads)
# Image processing and random distortion. Split across multiple threads
# with each thread applying a slightly different distortion.
assert self.config.num_preprocess_threads % 2 == 0
images_and_captions = []
for thread_id in range(self.config.num_preprocess_threads):
serialized_sequence_example = input_queue.dequeue()
encoded_image, caption = input_ops.parse_sequence_example(
serialized_sequence_example,
image_feature=self.config.image_feature_name,
caption_feature=self.config.caption_feature_name)
image = self.process_image(encoded_image, thread_id=thread_id)
images_and_captions.append([image, caption])
# Batch inputs.
queue_capacity = (2 * self.config.num_preprocess_threads *
self.config.batch_size)
images, input_seqs, target_seqs, input_mask = (
input_ops.batch_with_dynamic_pad(images_and_captions,
batch_size=self.config.batch_size,
queue_capacity=queue_capacity))
self.target_seqs = target_seqs
self.input_mask = input_mask
self.images = images
self.input_seqs = input_seqs
def build_inputs(self):
"""Input prefetching, preprocessing and batching.
Outputs:
self.images
self.input_seqs
self.target_seqs (training and eval only)
self.input_mask (training and eval only)
"""
if self.mode == "inference":
# In inference mode, images and inputs are fed via placeholders.
image_feed = tf.placeholder(dtype=tf.string, shape=[], name="image_feed")
input_feed = tf.placeholder(dtype=tf.int64,
shape=[None], # batch_size
name="input_feed")
# Process image and insert batch dimensions.
images = tf.expand_dims(self.process_image(image_feed), 0)
input_seqs = tf.expand_dims(input_feed, 1)
# No target sequences or input mask in inference mode.
target_seqs = None
input_mask = None
else:
# Prefetch serialized SequenceExample protos.
input_queue = input_ops.prefetch_input_data(
self.reader,
self.config.input_file_pattern,
is_training=self.is_training(),
batch_size=self.config.batch_size,
values_per_shard=self.config.values_per_input_shard,
input_queue_capacity_factor=self.config.input_queue_capacity_factor,
num_reader_threads=self.config.num_input_reader_threads)
# Image processing and random distortion. Split across multiple threads
# with each thread applying a slightly different distortion.
assert self.config.num_preprocess_threads % 2 == 0
images_and_objects_and_captions = []
for thread_id in range(self.config.num_preprocess_threads):
serialized_sequence_example = input_queue.dequeue()
encoded_image,object_feature, caption = input_ops.parse_sequence_example(
serialized_sequence_example,
image_feature=self.config.image_feature_name,
object_feature= self.config.image_object_feature_name,
caption_feature=self.config.caption_feature_name)
image = self.process_image(encoded_image, thread_id=thread_id)
images_and_objects_and_captions.append([image,object_feature, caption])
# Batch inputs.
queue_capacity = (2 * self.config.num_preprocess_threads *
self.config.batch_size)
images,objects_vector, input_seqs, target_seqs, input_mask = (
input_ops.batch_with_dynamic_pad(images_and_objects_and_captions,
batch_size=self.config.batch_size,
queue_capacity=queue_capacity))
self.images = images
self.objects_vector = objects_vector
self.input_seqs = input_seqs
self.target_seqs = target_seqs
self.input_mask = input_mask