def train_advanced_model(weight=None, batch_size=32, epochs=10):
cg = caption_generator.CaptionGenerator()
model = cg.create_advanced_model()
if weight != None:
model.load_weights(weight)
counter = 0
file_name = 'weights-improvement-{epoch:02d}.hdf5'
checkpoint = ModelCheckpoint(file_name,
monitor='loss',
verbose=1,
save_best_only=True,
mode='min')
callbacks_list = [checkpoint]
model.fit_generator(cg.data_generator(batch_size=batch_size),
steps_per_epoch=cg.total_samples / batch_size,
epochs=epochs,
verbose=2,
callbacks=callbacks_list)
try:
model.save('Models/WholeModel.h5', overwrite=True)
model.save_weights('Models/Weights.h5', overwrite=True)
except:
print "Error in saving model."
print "Training complete...\n"
def train_model(weight=None, batch_size=32, epochs=10):
cg = caption_generator.CaptionGenerator()
model = cg.create_model()
if weight != None:
model.load_weights(weight)
counter = 0
file_name = 'weights-improvement-epoch-{epoch:02d}-val_acc-{val_acc:.5f}-val_loss-{val_loss:.5f}.hdf5'
checkpoint = ModelCheckpoint(file_name,
monitor='val_acc',
verbose=1,
save_best_only=True,
mode='max')
callbacks_list = [checkpoint]
model.fit_generator(cg.data_generator(
path='Flickr8k_text/flickr_8k_train_dataset.txt',
batch_size=batch_size),
steps_per_epoch=cg.total_samples / batch_size,
epochs=epochs,
verbose=2,
callbacks=callbacks_list,
validation_steps=cg.total_samples_dev / batch_size,
validation_data=cg.data_generator(
path='Flickr8k_text/flickr_8k_dev_dataset.txt',
batch_size=batch_size))
try:
model.save('Models/WholeModel.h5', overwrite=True)
model.save_weights('Models/Weights.h5', overwrite=True)
except:
print "Error in saving model."
print "Training complete...\n"
def train_model(weight=None, batch_size=00, epochs=25):
# Total samples : 64146
cg = caption_generator.CaptionGenerator()
model = cg.create_model()
if weight != None:
model.load_weights(weight)
counter = 0
file_name = 'weights-improvement-{epoch:02d}.hdf5'
checkpoint = ModelCheckpoint(file_name,
monitor='loss',
verbose=1,
save_best_only=True,
mode='min')
callbacks_list = [checkpoint]
model.fit_generator(cg.data_generator(batch_size=batch_size),
steps_per_epoch=300,
epochs=epochs,
verbose=2,
callbacks=callbacks_list)
try:
model.save(
'C:/Users/pc/Desktop/automatic_image_caption_generation/Models/WholeModel.h5',
overwrite=True)
model.save_weights(
'C:/Users/pc/Desktop/automatic_image_caption_generation/Models/Weights.h5',
overwrite=True)
except:
print "Error in saving model."
print "Training complete...\n"
def _assertExpectedCaptions(self,
expected_captions,
beam_size=3,
max_caption_length=20,
length_normalization_factor=0):
"""Tests that beam search generates the expected captions.
Args:
expected_captions: A sequence of pairs (sentence, probability), where
sentence is a list of integer ids and probability is a float in [0, 1].
beam_size: Parameter passed to beam_search().
max_caption_length: Parameter passed to beam_search().
length_normalization_factor: Parameter passed to beam_search().
"""
expected_sentences = [c[0] for c in expected_captions]
expected_probabilities = [c[1] for c in expected_captions]
# Generate captions.
generator = caption_generator.CaptionGenerator(
model=FakeModel(),
vocab=FakeVocab(),
beam_size=beam_size,
max_caption_length=max_caption_length,
length_normalization_factor=length_normalization_factor)
actual_captions = generator.beam_search(sess=None, encoded_image=None)
actual_sentences = [c.sentence for c in actual_captions]
actual_probabilities = [math.exp(c.logprob) for c in actual_captions]
self.assertEqual(expected_sentences, actual_sentences)
self.assertAllClose(expected_probabilities, actual_probabilities)
def train_model(weight=None, batch_size=32, epochs=10):
cg = caption_generator.CaptionGenerator()
model = cg.create_model()
if weight is not None:
model.load_weights(weight)
file_name = 'weights-improvement-{epoch:02d}.hdf5'
checkpoint = ModelCheckpoint(file_name,
monitor='loss',
verbose=1,
save_best_only=True,
mode='min')
callbacks_list = [checkpoint]
optimizer = RMSprop(lr=0.001, rho=0.9, epsilon=1e-06)
parallel_model = multi_gpu_model(model, gpus=2)
parallel_model.compile(loss='categorical_crossentropy',
optimizer=optimizer,
metrics=['accuracy'])
parallel_model.fit_generator(cg.data_generator(batch_size=batch_size),
steps_per_epoch=cg.total_samples / batch_size,
epochs=epochs,
verbose=2,
callbacks=callbacks_list)
try:
parallel_model.save(os.path.join(CUR_DIR, '../Models/WholeModel.h5'),
overwrite=True)
parallel_model.save_weights(os.path.join(CUR_DIR,
'../Models/Weights.h5'),
overwrite=True)
except:
print("Error in saving model.")
print("Training complete...\n")
def train_model(weight = None, batch_size=32, epochs = 10):
cg = caption_generator.CaptionGenerator()
model = cg.create_model()
if weight != None:
model.load_weights(weight)
counter = 0
file_name = DATA_PATH + 'weights-checkpoint.hdf5'
#define callbacks
checkpoint = ModelCheckpoint(file_name, monitor='loss', verbose=1, save_best_only=True, mode='min')
tensor_board = TensorBoard(log_dir='./logs', write_graph=False, write_images=False)
reduce_lr = ReduceLROnPlateau(monitor='loss', factor=0.1, patience=8, verbose=1, min_lr=1e-6)
early_stopping = EarlyStopping(monitor='loss', min_delta=0.01, patience=16, verbose=1)
callbacks_list = [checkpoint, tensor_board, reduce_lr, early_stopping]
hist = model.fit_generator(cg.data_generator(batch_size=batch_size), steps_per_epoch=cg.total_samples/batch_size, epochs=epochs, verbose=2, callbacks=callbacks_list)
try:
model.save(DATA_PATH + 'final_model.h5', overwrite=True)
model.save_weights(DATA_PATH + 'final_weights.h5',overwrite=True)
except:
print "Error in saving model."
print "Training complete...\n"
return hist
def train_model(weight=None, batch_size=256, epochs=10):
cg = caption_generator.CaptionGenerator()
model = cg.create_model()
if weight != None:
model.load_weights(weight)
counter = 0
file_name = DATA_PATH + 'weights-checkpoint.h5'
#define callbacks
checkpoint = ModelCheckpoint(file_name,
monitor='val_loss',
verbose=1,
save_best_only=True,
mode='min')
tensor_board = TensorBoard(log_dir='./logs', write_graph=True)
hist_lr = LR_hist()
reduce_lr = LearningRateScheduler(step_decay)
early_stopping = EarlyStopping(monitor='val_loss',
min_delta=0.1,
patience=16,
verbose=1)
callbacks_list = [
checkpoint, tensor_board, hist_lr, reduce_lr, early_stopping
]
hist = model.fit_generator(
cg.data_generator_train(batch_size=batch_size),
steps_per_epoch=cg.total_samples / batch_size,
epochs=epochs,
verbose=2,
callbacks=callbacks_list,
validation_data=cg.data_generator_val(batch_size=batch_size),
validation_steps=cg.total_samples / (batch_size * 13.0))
model.save(DATA_PATH + 'final_model.h5', overwrite=True)
model.save_weights(DATA_PATH + 'final_weights.h5', overwrite=True)
hist_file = DATA_PATH + '/hist_model.dat'
with open(hist_file, 'w') as f:
pickle.dump(hist.history, f)
print "training complete...\n"
return model, hist, hist_lr
def main(_):
model = show_and_tell_model.ShowAndTellModel( FLAGS.model_path )
vocabi = vocab.Vocabulary(FLAGS.vocab_file)
filenames = _load_filenames()
generator = caption_generator.CaptionGenerator (model, vocabi)
for filename in filenames:
with tf.gfile.GFile(filename, "rb") as f:
image = f.read()
captions = generator.beam_search(image)
print("Captions for image %s:" % os.path.basename(filename))
for i, caption in enumerate(captions):
# Ignore begin and end tokens <S> and </S>.
sentence = [ vocab.Vocabulary.id_to_token(vocabi,w) for w in caption.sentence[1:-1]]
sentence = " ".join(sentence)
print(" %d) %s (p=%f)" % (i, sentence, math.exp(caption.logprob)))
def main(_):
# Build the inference graph.
g = tf.Graph()
with g.as_default():
model = inference_wrapper.InferenceWrapper()
restore_fn = model.build_graph_from_config(configuration.ModelConfig(),
FLAGS.checkpoint_path)
g.finalize()
# Create the vocabulary.
vocab = vocabulary.Vocabulary(FLAGS.vocab_file)
filenames = []
for file_pattern in FLAGS.input_files.split(","):
filenames.extend(tf.gfile.Glob(file_pattern))
tf.logging.info("Running caption generation on %d files matching %s",
len(filenames), FLAGS.input_files)
with tf.Session(graph=g) as sess:
# Load the model from checkpoint.
restore_fn(sess)
# Prepare the caption generator. Here we are implicitly using the default
# beam search parameters. See caption_generator.py for a description of the
# available beam search parameters.
generator = caption_generator.CaptionGenerator(model, vocab)
images = []
for filename in filenames:
with tf.gfile.GFile(filename, "rb") as f:
image = f.read()
images.append(image)
captions = generator.beam_search(sess, images, vocab)
for i, image in enumerate(images):
print("Captions for image %s:" % os.path.basename(filenames[i]))
for j, caption in enumerate(captions[i]):
# Ignore begin and end words.
sentence = [
vocab.id_to_word(w) for w in caption.sentence[1:-1]
]
sentence = " ".join(sentence)
print(" %d) %s (p=%f)" %
(j, sentence, math.exp(caption.logprob)))
def img_captions(file_inputs):
# Build the inference graph.
g = tf.Graph()
with g.as_default():
model = inference_wrapper.InferenceWrapper()
restore_fn = model.build_graph_from_config(configuration.ModelConfig(),
file_inputs[0])
g.finalize()
# Create the vocabulary.
vocab = vocabulary.Vocabulary(file_inputs[1])
filenames = []
for file_pattern in file_inputs[2].split(","):
filenames.extend(tf.gfile.Glob(file_pattern))
tf.logging.info("Running caption generation on %d files matching %s",
len(filenames), file_inputs[2])
with tf.Session(graph=g) as sess:
# Load the model from checkpoint.
restore_fn(sess)
# Prepare the caption generator. Here we are implicitly using the default
# beam search parameters. See caption_generator.py for a description of the
# available beam search parameters.
generator = caption_generator.CaptionGenerator(model, vocab)
caption_list = list()
prob_list = list()
for filename in filenames:
with tf.gfile.GFile(filename, "rb") as f:
image = f.read()
captions, probs = generator.beam_search(sess, image)
prob_list.append('['+", ".join(map(str, probs))+']')
loc_cap_list = list()
for i, caption in enumerate(captions):
# Ignore begin and end words.
sentence = [vocab.id_to_word(w) for w in caption.sentence[1:-1]]
sentence = " ".join(sentence).split('<S>')[0]
loc_cap_list.append([sentence, math.exp(caption.logprob)])
caption_list.append(loc_cap_list)
return prob_list, caption_list
def start_testing_images(model_path):
#Load model
detection_graph = load_frozen_model(model_path)
vocab = vocabulary.Vocabulary()
generator = caption_generator.CaptionGenerator(None, vocab)
with detection_graph.as_default():
with tf.Session(graph=detection_graph) as sess:
#print ([n.name for n in tf.get_default_graph().as_graph_def().node])
#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")
#for op in detection_graph.get_operations():
# print(op.name)
for image_path in TEST_IMAGE_PATHS:
start = timeit.default_timer()
image = tf.gfile.GFile(image_path, 'rb').read()
img = Image.open(image_path)
npimg = load_image_into_numpy_array(img)
print(type(image))
stop = timeit.default_timer()
print("Time to encode image: ", stop - start)
# Actual detection
start = timeit.default_timer()
captions = generator.beam_search(sess, image, img)
stop = timeit.default_timer()
print("Time to Generate captions: ", stop - start)
#Caption priting...
start = timeit.default_timer()
for i, caption in enumerate(captions):
sentence = [
vocab.id_to_word(w) for w in caption.sentence[1:-1]
]
sentence = " ".join(sentence)
print(" %d) %s (p=%f)" %
(i, sentence, math.exp(caption.logprob)))
stop = timeit.default_timer()
print("Time for Caption -> Sentence", stop - start)
def train_model(weight = None, batch_size=256, epochs = 10):
cg = caption_generator.CaptionGenerator()
model = cg.create_model()
model.summary()
if weight != None:
model.load_weights(weight)
counter = 0
file_name = 'weights-improvement-{epoch:02d}-{count:02d}.hdf5'
checkpoint = ModelCheckpoint(file_name, monitor='loss', verbose=1, save_best_only=True, mode='min')
callbacks_list = [checkpoint]
model.fit_generator(cg.data_generator(batch_size=batch_size), steps_per_epoch=cg.total_samples/batch_size, epochs=5, verbose=1, callbacks=callbacks_list)
#model.fit_generator(cg.data_generator(batch_size=batch_size), epochs=1, verbose=1, callbacks=callbacks_list)
counter = 1
model.optimizer.lr = 0.008
model.fit_generator(cg.data_generator(batch_size=batch_size), steps_per_epoch=cg.total_samples/batch_size, epochs=5, verbose=1, callbacks=callbacks_list)
counter = 2
model.optimizer.lr = 0.006
model.fit_generator(cg.data_generator(batch_size=batch_size), steps_per_epoch=cg.total_samples/batch_size, epochs=5, verbose=1, callbacks=callbacks_list)
counter = 4
model.optimizer.lr = 0.004
model.fit_generator(cg.data_generator(batch_size=batch_size), steps_per_epoch=cg.total_samples/batch_size, epochs=5, verbose=1, callbacks=callbacks_list)
counter = 5
model.optimizer.lr = 0.002
model.fit_generator(cg.data_generator(batch_size=batch_size), steps_per_epoch=cg.total_samples/batch_size, epochs=5, verbose=1, callbacks=callbacks_list)
counter = 6
model.optimizer.lr = 0.001
model.fit_generator(cg.data_generator(batch_size=batch_size), steps_per_epoch=cg.total_samples/batch_size, epochs=45, verbose=1, callbacks=callbacks_list)
try:
model.save('/home/manish.singhal/Image-Captioning-master/caption_generator/Models/WholeModel.h5', overwrite=True)
model.save_weights('/home/manish.singhal/Image-Captioning-master/caption_generator/Models/Weights.h5',overwrite=True)
except:
print ("Error in saving model.")
print ("Training complete...\n")
encoded_images = file['valid_set']
valid_list_file = "data/valid_list.txt"
train_step = conf.train_step
checkpoint_steps = conf.original_train_steps + (train_step -
1) * conf.interval_train_steps
check_point_path = "train_log/{}.ckpt".format(checkpoint_steps)
model = inference_wrapper.InferenceWrapper()
restore_fn = model.build_graph_from_config(configuration.ModelConfig(),
check_point_path)
sess = tf.InteractiveSession()
restore_fn(sess)
generator = caption_generator.CaptionGenerator(
model, vocab, beam_size=1, use_ngram=conf.use_ngram_gen_result)
valid_list_file = open(valid_list_file, 'r')
valid_image_list = []
for line in valid_list_file.readlines():
valid_image_list.append(line.strip().split()[0])
# output three optional sentences for each image, ranking by probability in decreasing order
# with open('/home/chengcheng/dataset/image_caption/inference/3/valid_caption_{}.txt'.format(checkpoint_steps), 'w') as f:
result_list = []
for index in range(1000):
captions = generator.beam_search(sess, encoded_images[index])
# if encoded_images[index] != valid_image_list[index]:
# print(encoded_images[index], valid_image_list[index])
if index % 100 == 0:
import cPickle as pickle
import caption_generator
import numpy as np
from keras.preprocessing import sequence
import nltk
cg = caption_generator.CaptionGenerator()
def process_caption(caption):
caption_split = caption.split()
processed_caption = caption_split[1:]
try:
end_index = processed_caption.index('<end>')
processed_caption = processed_caption[:end_index]
except:
pass
return " ".join([word for word in processed_caption])
def get_best_caption(captions):
captions.sort(key=lambda l: l[1])
best_caption = captions[-1][0]
return " ".join([cg.index_word[index] for index in best_caption])
def get_all_captions(captions):
final_captions = []
captions.sort(key=lambda l: l[1])
for caption in captions:
text_caption = " ".join([cg.index_word[index] for index in caption[0]])
import math
vocab = vocabulary.Vocabulary("data/dic.txt")
file = h5py.File("/home/chengcheng/dataset/image_caption/feat.hdf5", 'r')
encoded_images = file['valid_set']
valid_list_file = "/home/chengcheng/ImageCaption/data/valid_list.txt"
check_point_steps = 800000
model = inference_wrapper.InferenceWrapper()
restore_fn = model.build_graph_from_config(
configuration.ModelConfig(), "train_log/{}.ckpt".format(check_point_steps))
sess = tf.InteractiveSession()
restore_fn(sess)
generator = caption_generator.CaptionGenerator(model, vocab)
valid_list_file = open(valid_list_file, 'r')
valid_image_list = []
for line in valid_list_file.readlines():
valid_image_list.append(line.strip().split()[0])
# output three optional sentences for each image, ranking by probability in decreasing order
with open('infer_result/6000_0_valid_caption_{}.txt'.format(check_point_steps),
'w') as f:
for index in range(1000):
captions = generator.beam_search(sess, encoded_images[index])
f.write(valid_image_list[index])
# print("Captions for image {}".format(valid_image_list[index]))
for i, caption in enumerate(captions):
# Ignore begin and end words.
sentence = [vocab.id_to_word(w) for w in caption.sentence[1:-1]]
def main(_):
# Build the inference graph.
g = tf.Graph()
with g.as_default():
model = inference_wrapper.InferenceWrapper()
restore_fn = model.build_graph_from_config(configuration.ModelConfig(),
FLAGS.checkpoint_path)
g.finalize()
# Create the vocabulary.
vocab = vocabulary.Vocabulary(FLAGS.vocab_file)
filenames = []
for file_pattern in FLAGS.input_files.split(","):
filenames.extend(tf.gfile.Glob(file_pattern))
tf.logging.info("Running caption generation on %d files matching %s",
len(filenames), FLAGS.input_files)
with tf.Session(graph=g) as sess:
# Load the model from checkpoint.
restore_fn(sess)
# Prepare the caption generator. Here we are implicitly using the default
# beam search parameters. See caption_generator.py for a description of the
# available beam search parameters.
generator = caption_generator.CaptionGenerator(model, vocab)
##################
results = []
results0 = []
results1 = []
results2 = []
count = 1
for filename in filenames:
with tf.gfile.GFile(filename, "r") as f:
image = f.read()
captions = generator.beam_search(sess, image)
image_name_full = os.path.basename(filename)
print("Captions for %d/30000 image %s:" % (count, image_name_full))
count = count + 1
b = re.compile(r'.jpg')
image_name = b.sub('', image_name_full)
for i, caption in enumerate(captions):
# Ignore begin and end words.
sentence = [
vocab.id_to_word(w) for w in caption.sentence[1:-1]
]
sentence = "".join(sentence)
if i == 0:
results0.append({
"image_id": image_name,
"caption": sentence,
})
if i == 1:
results1.append({
"image_id": image_name,
"caption": sentence,
})
if i == 2:
results2.append({
"image_id": image_name,
"caption": sentence,
})
results.append({
"image_id": image_name,
"caption": sentence,
})
print(" %d) %s (p=%f)" %
(i, sentence, math.exp(caption.logprob)))
print("the length of results is:", len(results))
print("the length of results is:", len(results0))
print("the length of results is:", len(results1))
print("the length of results is:", len(results2))
#
outfile = "/media/han/6f586f18-792a-40fd-ada6-59702fb5dabc/wen/im2txt-aic/eval/data/val_results.json"
outfile0 = "/media/han/6f586f18-792a-40fd-ada6-59702fb5dabc/wen/im2txt-aic/eval/data/val_results0.json"
outfile1 = "/media/han/6f586f18-792a-40fd-ada6-59702fb5dabc/wen/im2txt-aic/eval/data/val_results1.json"
outfile2 = "/media/han/6f586f18-792a-40fd-ada6-59702fb5dabc/wen/im2txt-aic/eval/data/val_results2.json"
with io.open(outfile, 'w', encoding='utf-8') as fd:
fd.write(
unicode(
json.dumps(results,
ensure_ascii=False,
sort_keys=True,
indent=2,
separators=(',', ': '))))
with io.open(outfile0, 'w', encoding='utf-8') as fd0:
fd0.write(
unicode(
json.dumps(results0,
ensure_ascii=False,
sort_keys=True,
indent=2,
separators=(',', ': '))))
with io.open(outfile1, 'w', encoding='utf-8') as fd1:
fd1.write(
unicode(
json.dumps(results1,
ensure_ascii=False,
sort_keys=True,
indent=2,
separators=(',', ': '))))
with io.open(outfile2, 'w', encoding='utf-8') as fd2:
fd2.write(
unicode(
json.dumps(results2,
ensure_ascii=False,
sort_keys=True,
indent=2,
separators=(',', ': '))))
def caption_image():
"""API to caption images"""
image_format = "not jpeg"
st = current_time()
# get beam_size
beam_size = int(request.args.get("beam_size", "3"))
# get max_caption_length
max_caption_length = int(request.args.get("max_caption_length", "20"))
# get image_data
if request.method == 'POST':
image_data = request.get_data()
else:
url = request.args.get("url")
c_type, image_data = get_remote_file(url)
if not image_data:
return Response(status=400, response=jsonify(error="Could not HTTP GET %s" % url))
if 'image/jpeg' in c_type:
image_format = "jpeg"
# use c_type to find whether image_format is jpeg or not
# if jpeg, don't convert
if image_format == "jpeg":
jpg_image = image_data
# if not jpeg
else:
# open the image from raw bytes
image = Image.open(BytesIO(image_data))
# convert the image to RGB format, otherwise will give errors when converting to jpeg, if the image isn't RGB
rgb_image = image.convert("RGB")
# convert the RGB image to jpeg
image_bytes = BytesIO()
rgb_image.save(image_bytes, format="jpeg", quality=95)
jpg_image = image_bytes.getvalue()
image_bytes.close()
read_time = current_time() - st
# restart counter
st = current_time()
generator = caption_generator.CaptionGenerator(app.model,
app.vocab,
beam_size=beam_size,
max_caption_length=max_caption_length)
captions = generator.beam_search(app.sess, jpg_image)
captioning_time = current_time() - st
app.logger.info("Captioning time : %d" % captioning_time)
array_captions = []
for caption in captions:
sentence = [app.vocab.id_to_word(w) for w in caption.sentence[1:-1]]
sentence = " ".join(sentence)
array_captions.append({
'sentence': sentence,
'confidence': math.exp(caption.logprob)
})
response = {
'beam_size': beam_size,
'max_caption_length': max_caption_length,
'captions': array_captions,
'time': {
'read': read_time,
'captioning': captioning_time,
'units': 'ms'
}
}
return Response(response=json.dumps(response), status=200, mimetype="application/json")
def main(_):
#convert jpg image(s) into iamge representations using alexnet:
filenames = [
os.path.join(image_dir, f) for f in [
'overly-attached-girlfriend.jpg',
'high-expectations-asian-father.jpg', 'foul-bachelor-frog.jpg',
'stoner-stanley.jpg', 'y-u-no.jpg', 'willy-wonka.jpg',
'futurama-fry.jpg', 'success-kid.jpg', 'one-does-not-simply.jpg',
'bad-luck-brian.jpg', 'first-world-problems.jpg',
'philosoraptor.jpg', 'what-if-i-told-you.jpg', 'TutorPP.jpg'
]
]
print(filenames)
tf.logging.info("Running caption generation on %d files matching %s",
len(filenames), FLAGS.input_files)
#mean of imagenet dataset in BGR
imagenet_mean = np.array([104., 117., 124.], dtype=np.float32)
#placeholder for input and dropout rate
x_Alex = tf.placeholder(tf.float32, [1, 227, 227, 3])
keep_prob_Alex = tf.placeholder(tf.float32)
#create model with default config ( == no skip_layer and 1000 units in the last layer)
modelAlex = AlexNet(x_Alex, keep_prob_Alex, 1000, [], ['fc7', 'fc8'],
512) #maybe need to put fc8 in skip_layers
#define activation of last layer as score
score = modelAlex.fc6
meme_embeddings = []
with tf.Session() as sess:
# Initialize all variables
sess.run(tf.global_variables_initializer())
# Load the pretrained weights into the model
modelAlex.load_initial_weights(sess)
for i, meme in enumerate(filenames):
img = Image.open(meme)
try:
img.thumbnail((227, 227), Image.ANTIALIAS)
#img = img.resize((227,227))
#use img.thumbnail for square images, img.resize for non square
assert np.shape(img) == (227, 227, 3)
except AssertionError:
img = img.resize((227, 227))
print('sizing error')
# Subtract the ImageNet mean
img = img - imagenet_mean #should probably change this
# Reshape as needed to feed into model
img = img.reshape((1, 227, 227, 3))
meme_vector = sess.run(score,
feed_dict={
x_Alex: img,
keep_prob_Alex: 1
}) #[1,4096]
meme_vector = np.reshape(meme_vector, [4096])
assert np.shape(meme_vector) == (4096, )
#now have np embeddings to feed for inference
meme_embeddings.append(meme_vector)
with open('Captions.txt', 'r') as f:
data_captions = f.readlines()
data_captions = [s.lower() for s in data_captions]
# Build the inference graph.
g = tf.Graph()
with g.as_default():
model = inference_wrapper.InferenceWrapper()
restore_fn = model.build_graph_from_config(configuration.ModelConfig(),
FLAGS.checkpoint_path)
g.finalize()
# Create the vocabulary.
vocab = vocabulary.Vocabulary(FLAGS.vocab_file)
#filenames = []
#for file_pattern in FLAGS.input_files.split(","):
#filenames.extend(tf.gfile.Glob(file_pattern))
#tf.logging.info("Running caption generation on %d files matching %s",
#len(filenames), FLAGS.input_files)
with tf.Session(graph=g) as sess:
# Load the model from checkpoint.
restore_fn(sess)
# Prepare the caption generator. Here we are implicitly using the default
# beam search parameters. See caption_generator.py for a description of the
# available beam search parameters.
generator = caption_generator.CaptionGenerator(model, vocab)
num_in_data_total = 0
num_captions = 0
for i, meme in enumerate(meme_embeddings):
#with tf.gfile.GFile(filename, "rb") as f:
#image = f.read()
captions = generator.beam_search(sess, meme)
print("Captions for image %s:" % os.path.basename(filenames[i]))
num_in_data = 0
for i, caption in enumerate(captions):
# Ignore begin and end words.
sentence = [
vocab.id_to_word(w) for w in caption.sentence[1:-1]
]
sentence = " ".join(sentence)
in_data = 0
if b_any(sentence in capt for capt in data_captions):
in_data = 1
num_in_data += 1
num_in_data_total += 1
num_captions += 1
else:
num_captions += 1
print(" %d) %s (p=%f) [in data = %d]" %
(i, sentence, math.exp(caption.logprob), in_data))
print("number of captions in data = %d" % (num_in_data))
print("(total number of captions in data = %d) percent in data = %f" %
(num_in_data_total, (num_in_data_total / num_captions)))
train_vector_file = "data/train_vector.txt"
train_step = conf.train_step
checkpoint_steps = conf.original_train_steps + (train_step -
1) * conf.interval_train_steps
checkpoint_path = "train_log/{}.ckpt".format(checkpoint_steps)
model = inference_wrapper.InferenceWrapper()
restore_fn = model.build_graph_from_config(configuration.ModelConfig(),
checkpoint_path)
sess = tf.InteractiveSession()
restore_fn(sess)
generator = caption_generator.CaptionGenerator(
model, vocab, beam_size=conf.beam_size, use_ngram=conf.use_ngram_gen_label)
train_list_file = open(train_list_file, 'r')
train_image_list = []
for line in train_list_file.readlines():
train_image_list.append(line.strip().split()[0])
# output three optional sentences for each image, ranking by probability in decreasing order
# with open('/home/chengcheng/dataset/image_caption/inference/3/train_caption_{}.txt'.format(check_point_steps), 'w') as f:
caption_vector_path = "train_log/{}_infer_train_vector.txt".format(
checkpoint_steps)
label_file = open(caption_vector_path, 'w')
index = -1
for line in open(train_vector_file):