def run():
"""Runs evaluation in a loop, and logs summaries to TensorBoard."""
# Create the evaluation directory if it doesn't exist.
eval_dir = FLAGS.eval_dir
if not tf.gfile.IsDirectory(eval_dir):
tf.logging.info("Creating eval directory: %s", eval_dir)
tf.gfile.MakeDirs(eval_dir)
g = tf.Graph()
with g.as_default():
# Build the model for evaluation.
model_config = configuration.ModelConfig()
model_config.input_file_pattern = FLAGS.input_file_pattern
model = show_and_tell_model.ShowAndTellModel(model_config, mode="eval")
model.build()
# Create the summary operation and the summary writer.
summary_op = tf.merge_all_summaries()
summary_writer = tf.train.SummaryWriter(eval_dir)
g.finalize()
# Run a new evaluation run every eval_interval_secs.
while True:
start = time.time()
tf.logging.info(
"Starting evaluation at " +
time.strftime("%Y-%m-%d-%H:%M:%S", time.localtime()))
run_once(model, summary_writer, summary_op)
time_to_next_eval = start + FLAGS.eval_interval_secs - time.time()
if time_to_next_eval > 0:
time.sleep(time_to_next_eval)
def build_model(self,
model_config,
image_raw_feed=None,
input_feed=None,
mask_feed=None):
model = show_and_tell_model.ShowAndTellModel(model_config,
mode="attack")
model.build(image_raw_feed, input_feed, mask_feed)
self.model = model
return model
def run_caption(checkpoint_path, filenames, heuristic_amount):
g = tf.Graph()
with g.as_default():
# Build the model for evaluation.
model_config = configuration.ModelConfig()
model = show_and_tell_model.ShowAndTellModel(model_config,
mode="inference")
model.build()
# Create the Saver to restore model Variables.
saver = tf.train.Saver()
g.finalize()
def _restore_fn(sess):
tf.logging.info("Loading model from checkpoint: %s", checkpoint_path)
saver.restore(sess, checkpoint_path)
tf.logging.info("Successfully loaded checkpoint: %s",
os.path.basename(checkpoint_path))
restore_fn = _restore_fn
# Create the vocabulary.
vocab = glove.load(model_config.config)[0]
run_results = []
with tf.Session(graph=g) as sess:
# Load the model from checkpoint.
restore_fn(sess)
# Prepare the list of image bytes for evaluation.
images = []
for filename in filenames:
with tf.gfile.GFile(filename, "rb") as f:
images.append(f.read())
captions = [
sess.run(model.final_seqs,
feed_dict={
"image_feed:0": img,
model.heuristic_temperature: heuristic_amount
}) for img in images
]
for i, caption in enumerate(captions):
run_results.append({"filename": filenames[i], "captions": []})
for j in range(caption.shape[1]):
# Ignore begin and end words.
c = caption[0, j, :].tolist()
sentence = [vocab.id_to_word(w) for w in c[:-1]]
sentence = " ".join(sentence)
run_results[i]["captions"].append(sentence)
return run_results
def run():
"""Runs evaluation in a loop, and logs summaries to TensorBoard."""
# Create the evaluation directory if it doesn't exist.
eval_dir = FLAGS.eval_dir
if not tf.gfile.IsDirectory(eval_dir):
tf.logging.info("Creating eval directory: %s", eval_dir)
tf.gfile.MakeDirs(eval_dir)
g = tf.Graph()
with g.as_default():
# Build the model for evaluation.
model_config = configuration.ModelConfig()
model_config.input_file_pattern = FLAGS.input_file_pattern
model_config.CNN_name = FLAGS.CNN_name
# 若FLAGS.batch_size设置了，则以运行时的设置为准，否则以configuration.py中设置为准
if FLAGS.batch_size > 0:
model_config.batch_size = FLAGS.batch_size
model = show_and_tell_model.ShowAndTellModel(model_config, mode="eval")
model.build()
# Create the Saver to restore model Variables.
saver = tf.train.Saver()
# Create the summary operation and the summary writer.
summary_op = tf.summary.merge_all()
summary_writer = tf.summary.FileWriter(eval_dir)
g.finalize()
eval_once = (FLAGS.eval_interval_secs == 0)
# Run a new evaluation run every eval_interval_secs.
while True:
start = time.time()
tf.logging.info(
"Starting evaluation at " +
time.strftime("%Y-%m-%d-%H:%M:%S", time.localtime()))
run_once(model, saver, summary_writer, summary_op)
if eval_once:
tf.logging.info("Just eval once at this time!")
break
time_to_next_eval = start + FLAGS.eval_interval_secs - time.time()
if time_to_next_eval > 0:
time.sleep(time_to_next_eval)
def build_model(self, model_config):
model = show_and_tell_model.ShowAndTellModel(model_config,
mode="inference")
model.build()
return model
def main(unused_argv):
assert FLAGS.input_file_pattern, "--input_file_pattern is required"
assert FLAGS.train_dir, "--train_dir is required"
model_config = configuration.ModelConfig()
model_config.input_file_pattern = FLAGS.input_file_pattern
model_config.inception_checkpoint_file = FLAGS.inception_checkpoint_file
training_config = configuration.TrainingConfig()
# Create training directory.
train_dir = FLAGS.train_dir
if not tf.gfile.IsDirectory(train_dir):
tf.logging.info("Creating training directory: %s", train_dir)
tf.gfile.MakeDirs(train_dir)
# Build the TensorFlow graph.
g = tf.Graph()
with g.as_default():
# Build the model.
model = show_and_tell_model.ShowAndTellModel(
model_config, mode="train", train_inception=FLAGS.train_inception)
model.build()
# Set up the learning rate.
learning_rate_decay_fn = None
if FLAGS.train_inception:
learning_rate = tf.constant(
training_config.train_inception_learning_rate)
else:
learning_rate = tf.constant(training_config.initial_learning_rate)
if training_config.learning_rate_decay_factor > 0:
num_batches_per_epoch = (
training_config.num_examples_per_epoch /
model_config.batch_size)
decay_steps = int(num_batches_per_epoch *
training_config.num_epochs_per_decay)
def _learning_rate_decay_fn(learning_rate, global_step):
return tf.train.exponential_decay(
learning_rate,
global_step,
decay_steps=decay_steps,
decay_rate=training_config.learning_rate_decay_factor,
staircase=True)
learning_rate_decay_fn = _learning_rate_decay_fn
# Set up the training ops.
train_op = tf.contrib.layers.optimize_loss(
loss=model.total_loss,
global_step=model.global_step,
learning_rate=learning_rate,
optimizer=training_config.optimizer,
clip_gradients=training_config.clip_gradients,
learning_rate_decay_fn=learning_rate_decay_fn)
# Set up the Saver for saving and restoring model checkpoints.
saver = tf.train.Saver(
max_to_keep=training_config.max_checkpoints_to_keep)
# Run training.
tf.contrib.slim.learning.train(train_op,
train_dir,
log_every_n_steps=FLAGS.log_every_n_steps,
graph=g,
global_step=model.global_step,
number_of_steps=FLAGS.number_of_steps,
init_fn=model.init_fn,
saver=saver)
def build_model(self, model_config): model = show_and_tell_model.ShowAndTellModel(model_config, mode="saliency") model.config.batch_size = self.batch_size model.build() return model
def main(unused_argv):
assert FLAGS.input_file_pattern, "--input_file_pattern is required"
assert FLAGS.train_dir, "--train_dir is required"
model_config = configuration.ModelConfig()
model_config.input_file_pattern = FLAGS.input_file_pattern
model_config.image_keys = [model_config.image_feature_name]
#make sure we have the right batch size
if FLAGS.train_inception:
assert FLAGS.batch_size == 8
else:
assert FLAGS.batch_size == 32
if FLAGS.two_input_queues:
FLAGS.batch_size = int(FLAGS.batch_size / 2)
model_config.batch_size = FLAGS.batch_size
#assert all batch sizes are right
#set flags if you are training with blocked image
if FLAGS.blocked_image:
assert FLAGS.blocked_input_file_pattern, "--blocked_input_file_pattern is required if you would like to train with blocked images"
model_config.blocked_input_file_pattern = FLAGS.blocked_input_file_pattern
model_config.image_keys.append(model_config.blocked_image_feature_name)
if FLAGS.two_input_queues:
assert FLAGS.input_file_pattern2, "--input_file_pattern2 is required if you would like to train with two input queues"
model_config.blocked_input_file_pattern = FLAGS.input_file_pattern2
model_config.image_keys.append(model_config.image_feature_name)
model_config.inception_checkpoint_file = FLAGS.inception_checkpoint_file
training_config = configuration.TrainingConfig()
# Create training directory.
train_dir = FLAGS.train_dir
if not tf.gfile.IsDirectory(train_dir):
tf.logging.info("Creating training directory: %s", train_dir)
tf.gfile.MakeDirs(train_dir)
print('graph')
#go from flags to dict
g = tf.Graph()
with g.as_default():
# Build the model.
if not isinstance(
FLAGS.__flags['init_from'], str
): #Tensorflow likes to change random things for different releases. One random thing it likes to change is FLAGS. This code takes care of that *sight*
flag_dict = {}
for key in FLAGS.__flags.keys():
flag_dict[key] = FLAGS.__flags[key].value
else:
flag_dict = FLAGS.__flags
model = show_and_tell_model.ShowAndTellModel(
model_config,
mode="train",
train_inception=FLAGS.train_inception,
flags=flag_dict
) #let's just pass in all the flags bc this is going to get annoying
model.build()
# Set up the learning rate.
learning_rate_decay_fn = None
if FLAGS.train_inception:
learning_rate = tf.constant(
training_config.train_inception_learning_rate)
else:
learning_rate = tf.constant(training_config.initial_learning_rate)
if training_config.learning_rate_decay_factor > 0:
num_batches_per_epoch = (
training_config.num_examples_per_epoch /
model_config.batch_size)
decay_steps = int(num_batches_per_epoch *
training_config.num_epochs_per_decay)
def _learning_rate_decay_fn(learning_rate, global_step):
return tf.train.exponential_decay(
learning_rate,
global_step,
decay_steps=decay_steps,
decay_rate=training_config.learning_rate_decay_factor,
staircase=True)
learning_rate_decay_fn = _learning_rate_decay_fn
# Set up the training ops.
train_op = tf.contrib.layers.optimize_loss(
loss=model.total_loss,
global_step=model.global_step,
learning_rate=learning_rate,
optimizer=training_config.optimizer,
clip_gradients=training_config.clip_gradients,
learning_rate_decay_fn=learning_rate_decay_fn)
# Set up the Saver for saving and restoring model checkpoints.
saver = tf.train.Saver(
max_to_keep=training_config.max_checkpoints_to_keep)
if FLAGS.init_from:
inception_restore = model.init_fn
def restore_full_model(sess):
print("restoring full model")
inception_restore(sess)
saver.restore(sess, FLAGS.init_from)
model.init_fn = restore_full_model
print('train')
# Run training.
if FLAGS.debug:
tf.contrib.slim.learning.train(
train_op,
train_dir,
log_every_n_steps=FLAGS.log_every_n_steps,
graph=g,
global_step=model.global_step,
number_of_steps=FLAGS.number_of_steps,
init_fn=model.init_fn,
saver=saver,
session_wrapper=tf_debug.LocalCLIDebugWrapperSession)
else:
tf.contrib.slim.learning.train(
train_op,
train_dir,
log_every_n_steps=FLAGS.log_every_n_steps,
graph=g,
global_step=model.global_step,
number_of_steps=FLAGS.number_of_steps,
init_fn=model.init_fn,
saver=saver)
def run_caption(checkpoint_path, attrigram_checkpoint_path, filenames,
divisions):
g = tf.Graph()
with g.as_default():
# Build the model for evaluation.
model_config = configuration.ModelConfig()
model = show_and_tell_model.ShowAndTellModel(model_config,
mode="inference")
model.use_style = True
model.build()
# Create the Saver to restore model Variables.
saver = tf.train.Saver(var_list=model.model_variables)
g.finalize()
def _restore_fn(sess):
tf.logging.info("Loading model from checkpoint: %s", checkpoint_path)
saver.restore(sess, checkpoint_path)
tf.logging.info("Successfully loaded checkpoint: %s",
os.path.basename(checkpoint_path))
restore_fn = _restore_fn
# Create the vocabulary.
vocab = glove.load(model_config.config)[0]
run_results = []
with tf.Session(graph=g) as sess:
# Load the model from checkpoint.
restore_fn(sess)
# Run the attribute probabilities
ps = run_attributes.run_attributes(attrigram_checkpoint_path,
filenames, divisions)
# Prepare the list of image bytes for evaluation.
images = []
for filename in filenames:
with tf.gfile.GFile(filename, "rb") as f:
images.append(f.read())
captions = []
for img, p in zip(images, ps):
input_feed = {
"image_feed:0": img,
model.attribute_probabilities: [p["probabilities"]]
}
sess.run(model.assign_initial_states, feed_dict=input_feed)
# The original caption
img_progress = [sess.run(model.style_seqs, feed_dict=input_feed)]
# Perform style transfer
for x in range(10):
sess.run(model.descend_style, feed_dict=input_feed)
caption = sess.run(model.style_seqs, feed_dict=input_feed)
img_progress.append(caption)
captions.append(img_progress)
for i, caption in enumerate(captions):
run_results.append({"filename": filenames[i], "captions": []})
for flow in caption:
for j in range(flow.shape[0]):
# Ignore begin and end words.
c = flow[j, 0, :].tolist()
sentence = [vocab.id_to_word(w) for w in c[:-1]]
sentence = " ".join(sentence)
run_results[i]["captions"].append(sentence)
return run_results
tf.logging.set_verbosity(tf.logging.INFO)
assert FLAGS.input_file_pattern, "--input_file_pattern is required"
assert FLAGS.train_dir, "--train_dir is required"
model_config = configuration.ModelConfig()
model_config.input_file_pattern = FLAGS.input_file_pattern
model_config.inception_checkpoint_file = FLAGS.inception_checkpoint_file
training_config = configuration.TrainingConfig()
g = tf.Graph()
with g.as_default():
# build the model
model = show_and_tell_model.ShowAndTellModel(
model_config, mode='train', train_inception=FLAGS.train_inception)
model.build()
# set up the learning rate
learning_rate_decay_fn = None
if FLAGS.train_inception:
learning_rate = tf.constant(
training_config.train_inception_learning_rate)
else:
learning_rate = tf.constant(training_config.initial_learning_rate)
if training_config.learning_rate_decay_factor > 0:
num_batches_per_epoch = training_config.num_examples_per_epoch / model_config.batch_size
decay_steps = int(num_batches_per_epoch *
training_config.num_epochs_per_decay)
def train(number_of_steps):
model_config = configuration.ModelConfig()
model_config.input_file_pattern = FLAGS.input_file_pattern
training_config = configuration.TrainingConfig()
model_config.inception_checkpoint_file = FLAGS.inception_checkpoint_file
# Create training directory.
train_dir = FLAGS.train_dir
# if not tf.gfile.IsDirectory(train_dir):
# tf.logging.info("Creating training directory: %s", train_dir)
# tf.gfile.MakeDirs(train_dir)
# Build the TensorFlow graph.
g = tf.Graph()
with g.as_default():
# Build the model.
model = show_and_tell_model.ShowAndTellModel(
model_config, mode="train", train_inception=FLAGS.train_inception)
model.build()
# Set up the learning rate.
learning_rate_decay_fn = None
if FLAGS.train_inception:
print(
"The inception weights are fine-tuned together with weights in the LSTM units and word embeddings."
)
learning_rate = tf.constant(
training_config.train_inception_learning_rate)
else:
print(
"The inception weights are frozen. Only weights in the LSTMs and word embeddings are randomly"
"initialized and trained.")
learning_rate = tf.constant(training_config.initial_learning_rate)
if training_config.learning_rate_decay_factor > 0:
num_batches_per_epoch = (
training_config.num_examples_per_epoch /
model_config.batch_size)
decay_steps = int(num_batches_per_epoch *
training_config.num_epochs_per_decay)
def _learning_rate_decay_fn(learning_rate, global_step):
return tf.train.exponential_decay(
learning_rate,
global_step,
decay_steps=decay_steps,
decay_rate=training_config.learning_rate_decay_factor,
staircase=True)
learning_rate_decay_fn = _learning_rate_decay_fn
# Set up the training ops.
train_op = tf.contrib.layers.optimize_loss(
loss=model.total_loss,
global_step=model.global_step,
learning_rate=learning_rate,
optimizer=training_config.optimizer,
clip_gradients=training_config.clip_gradients,
learning_rate_decay_fn=learning_rate_decay_fn)
saver = tf.train.Saver(keep_checkpoint_every_n_hours=0.5)
# saver = tf.train.Saver(max_to_keep=training_config.max_checkpoints_to_keep)
# Run training.
tf.contrib.slim.learning.train(train_op,
train_dir,
log_every_n_steps=FLAGS.log_every_n_steps,
graph=g,
global_step=model.global_step,
number_of_steps=number_of_steps,
init_fn=model.init_fn,
saver=saver)
def build_model_eval(self, model_config):
model = show_and_tell_model.ShowAndTellModel(model_config, mode="eval")
model.build()
self.model = model
return model
def build_model(self, model_config):
model = show_and_tell_model.ShowAndTellModel(model_config,
mode="inference")
model.build()
self.global_stepp = model.global_step
return model
