def __init__(self, model_dir, class_type):
self.graph = tf.Graph()
# config = tf.ConfigProto(allow_soft_placement=True)
# self.sess = tf.Session(config=config, graph=self.graph)
self.sess = tf.Session(graph=self.graph)
with self.graph.as_default():
# import saved model from loc into local graph
label_list = AGE_LIST if class_type == 'age' else GENDER_LIST
nlabels = len(label_list)
if class_type == "age":
model_fn = select_model(FLAGS.age_model_type)
else:
model_fn = select_model(FLAGS.sex_model_type)
self.images = tf.placeholder(tf.float32,
[None, RESIZE_FINAL, RESIZE_FINAL, 3])
logits = model_fn(nlabels, self.images, 1, False)
# init = tf.global_variables_initializer()
saver = tf.train.Saver()
requested_step = FLAGS.requested_step if FLAGS.requested_step else None
checkpoint_path = '%s' % (model_dir)
model_checkpoint_path, global_step = get_checkpoint(
checkpoint_path, requested_step, FLAGS.checkpoint)
saver.restore(self.sess, model_checkpoint_path)
self.softmax_output = tf.nn.softmax(logits)
def evaluate(run_dir):
with tf.Graph().as_default() as g:
input_file = os.path.join(FLAGS.train_dir, 'md.json')
print(input_file)
with open(input_file, 'r') as f:
md = json.load(f)
eval_data = FLAGS.eval_data == 'valid'
num_eval = md['%s_counts' % FLAGS.eval_data]
model_fn = select_model(FLAGS.model_type)
with tf.device(FLAGS.device_id):
print('Executing on %s' % FLAGS.device_id)
images, labels, _ = inputs(FLAGS.train_dir, FLAGS.batch_size, FLAGS.image_size, train=not eval_data, num_preprocess_threads=FLAGS.num_preprocess_threads)
logits = model_fn(md['nlabels'], images, 1, False)
summary_op = tf.summary.merge_all()
summary_writer = tf.summary.FileWriter(run_dir, g)
saver = tf.train.Saver()
if FLAGS.requested_step_seq:
sequence = FLAGS.requested_step_seq.split(',')
for requested_step in sequence:
print('Running %s' % sequence)
eval_once(saver, summary_writer, summary_op, logits, labels, num_eval, requested_step)
else:
while True:
print('Running loop')
eval_once(saver, summary_writer, summary_op, logits, labels, num_eval)
if FLAGS.run_once:
break
time.sleep(FLAGS.eval_interval_secs)
def evaluate(run_dir):
with tf.Graph().as_default() as g:
input_file = os.path.join(FLAGS.train_dir, 'md.json')
#print(input_file)
with open(input_file, 'r') as f:
md = json.load(f)
eval_data = FLAGS.eval_data == 'valid'
num_eval = md['%s_counts' % FLAGS.eval_data]
model_fn = select_model(FLAGS.model_type)
with tf.device(FLAGS.device_id):
#print('Executing on %s' % FLAGS.device_id)
images, labels, _ = inputs(FLAGS.train_dir, FLAGS.batch_size, FLAGS.image_size, train=not eval_data, num_preprocess_threads=FLAGS.num_preprocess_threads)
logits = model_fn(md['nlabels'], images, 1, False)
summary_op = tf.summary.merge_all()
summary_writer = tf.summary.FileWriter(run_dir, g)
saver = tf.train.Saver()
if FLAGS.requested_step_seq:
sequence = FLAGS.requested_step_seq.split(',')
for requested_step in sequence:
#print('Running %s' % sequence)
eval_once(saver, summary_writer, summary_op, logits, labels, num_eval, requested_step)
else:
while True:
#print('Running loop')
eval_once(saver, summary_writer, summary_op, logits, labels, num_eval)
if FLAGS.run_once:
break
time.sleep(FLAGS.eval_interval_secs)
def __init__(
self,
model_dir='/usr/src/app/deps/rude-carnie/inception_gender_checkpoint',
model_type='inception',
class_type='gender'):
'''
Just a wrapper around guess.py.
'''
self.model_dir = model_dir
self.model_type = model_type
self.class_type = class_type
self.sess = tf.Session()
model_fn = select_model(self.model_type)
self.label_list = AGE_LIST if self.class_type == 'age' else GENDER_LIST
nlabels = len(self.label_list)
self.images = tf.placeholder(tf.string, [None])
standardize = tf.map_fn(self.decode, self.images, dtype=tf.float32)
logits = model_fn(nlabels, standardize, 1, False)
init = tf.global_variables_initializer()
requested_step = None #FLAGS.requested_step if FLAGS.requested_step else None
checkpoint_path = '%s' % (self.model_dir)
model_checkpoint_path, global_step = get_checkpoint(
checkpoint_path, requested_step, None)
#FLAGS.checkpoint)
saver = tf.train.Saver()
saver.restore(self.sess, model_checkpoint_path)
self.softmax_output = tf.nn.softmax(logits)
self.coder = ImageCoder()
def guess_loop(class_type, model_type, model_dir, q, guess_q):
config = tf.ConfigProto(allow_soft_placement=True)
with tf.Session(config=config) as sess:
label_list = AGE_LIST if class_type == 'age' else GENDER_LIST
nlabels = len(label_list)
model_fn = select_model(model_type)
with tf.device('/cpu:0'):
images = tf.placeholder(tf.float32, [None, RESIZE_FINAL, RESIZE_FINAL, 3])
logits = model_fn(nlabels, images, 1, False)
init = tf.global_variables_initializer()
checkpoint_path = '%s' % (model_dir)
model_checkpoint_path, global_step = get_checkpoint(checkpoint_path, None, 'checkpoint')
saver = tf.train.Saver()
saver.restore(sess, model_checkpoint_path)
softmax_output = tf.nn.softmax(logits)
while True:
image_files = q.get()
guess_result = []
for image in image_files:
guess = classify_one_multi_crop(sess, label_list, softmax_output, images, image)
if guess is not None:
guess_result.append(guess)
if guess_q.full() is not True:
guess_q.put(guess_result)
def model_prediction_view():
display_image('undraw_viral_tweet_gndb', '')
st.info("Classification with Machine Learning Models")
st.markdown("""
### Use some of the built-in models to find out the sentiment of your tweet.
#### **Instructions:**
1. Simply pick a model you would like used to classify your tweet from the dropdown menu
2. Type in your tweet on the text area
3. Click the 'Classify' button and see your results below.
""")
# Creating sidebar with selection box -
# you can create multiple pages this way
models = ("Logistic_regression", "Linear SVC", "SVM")
st.subheader('Pick a model to classify your text:')
chosen = st.selectbox('', models)
if chosen in models:
select_model(chosen)
def guessGender(path): # pylint: disable=unused-argument
with tf.Session() as sess:
# tf.reset_default_graph()
label_list = GENDER_LIST
nlabels = len(label_list)
print('Executing on %s' % FLAGS.device_id)
model_fn = select_model(FLAGS.model_type)
with tf.device(FLAGS.device_id):
images = tf.placeholder(tf.float32,
[None, RESIZE_FINAL, RESIZE_FINAL, 3])
logits = model_fn(nlabels, images, 1, False)
init = tf.global_variables_initializer()
requested_step = FLAGS.requested_step if FLAGS.requested_step else None
checkpoint_path = '%s' % (GENDER_MODEL_PATH)
model_checkpoint_path, global_step = get_checkpoint(
checkpoint_path, requested_step, FLAGS.checkpoint)
saver = tf.train.Saver()
saver.restore(sess, model_checkpoint_path)
softmax_output = tf.nn.softmax(logits)
coder = ImageCoder()
files = deal_file.get_files(path)
font = cv2.FONT_HERSHEY_SIMPLEX
try:
for f in files:
best_choice = classify(sess, label_list, softmax_output,
coder, images, f)
# print(best_choice)
pic = cv2.imread(f)
fname = f[:-4] + '_test' + f[-4:]
cv2.putText(pic, best_choice[0], (5, 40), font, 1,
(100, 255, 50), 2, cv2.LINE_AA)
cv2.imwrite(fname, pic)
print(best_choice)
except Exception as e:
print(e)
print('Failed to run image %s ' % file)
def run_inference_on_image(image_data):
config = tf.ConfigProto(allow_soft_placement=True)
tf.logging.set_verbosity(tf.logging.WARN)
with tf.Session(config=config) as sess:
softmax_output = sess.graph.get_tensor_by_name('labels_softmax:0')
images = sess.graph.get_tensor_by_name('Placeholder:0')
label_list = AGE_LIST
nlabels = len(label_list)
model_fn = select_model('inception_v3')
#print('Executing on %s' % FLAGS.device_id)
image_bytes = download_image(url)
coder = ImageCoder()
return classify_many_single_crop(sess, label_list, softmax_output,
coder, images, image_bytes)
def extract(self):
with tf.Session(config=self.config) as sess:
nlabels = len(self.label_list)
model_fn = select_model(FLAGS.model_type)
with tf.device(FLAGS.device_id):
logits = model_fn(nlabels, images, 1, False)
images = tf.placeholder(tf.float32, [None, RESIZE_FINAL, RESIZE_FINAL, 3])
init = tf.global_variables_initializer()
requested_step = FLAGS.requested_step if FLAGS.requested_step else None
checkpoint_path = '%s' % (FLAGS.model_dir)
model_checkpoint_path, global_step = get_checkpoint(checkpoint_path, requested_step, FLAGS.checkpoint)
saver = tf.train.Saver()
saver.restore(sess, model_checkpoint_path)
softmax_output = tf.nn.softmax(logits)
coder = ImageCoder()
# Support a batch mode if no face detection model
if len(files) == 0:
if (os.path.isdir(FLAGS.filename)):
for relpath in os.listdir(FLAGS.filename):
abspath = os.path.join(FLAGS.filename, relpath)
if os.path.isfile(abspath) and any([abspath.endswith('.' + ty) for ty in ('jpg', 'png', 'JPG', 'PNG', 'jpeg')]):
print(abspath)
files.append(abspath)
else:
files.append(FLAGS.filename)
# If it happens to be a list file, read the list and clobber the files
if any([FLAGS.filename.endswith('.' + ty) for ty in ('csv', 'tsv', 'txt')]):
files = list_images(FLAGS.filename)
writer = None
output = None
if FLAGS.target:
print('Creating output file %s' % FLAGS.target)
output = open(FLAGS.target, 'w')
writer = csv.writer(output)
writer.writerow(('file', 'label', 'score'))
image_files = list(filter(lambda x: x is not None, [resolve_file(f) for f in files]))
print(image_files)
if FLAGS.single_look:
classify_many_single_crop(sess, label_list, softmax_output, coder, images, image_files, writer)
else:
for image_file in image_files:
classify_one_multi_crop(sess, label_list, softmax_output, coder, images, image_file, writer)
if output is not None:
output.close()
def guessGender(path): # pylint: disable=unused-argument
# 检测文件夹中所有照片的性别
with tf.Session() as sess:
# tf.reset_default_graph()
label_list = GENDER_LIST
nlabels = len(label_list)
print('Executing on %s' % FLAGS.device_id)
model_fn = select_model(FLAGS.model_type)
with tf.device(FLAGS.device_id):
images = tf.placeholder(tf.float32, [None, RESIZE_FINAL, RESIZE_FINAL, 3])
logits = model_fn(nlabels, images, 1, False)
init = tf.global_variables_initializer()
requested_step = FLAGS.requested_step if FLAGS.requested_step else None
checkpoint_path = '%s' % (GENDER_MODEL_PATH)
model_checkpoint_path, global_step = get_checkpoint(checkpoint_path, requested_step, FLAGS.checkpoint)
saver = tf.train.Saver()
saver.restore(sess, model_checkpoint_path)
softmax_output = tf.nn.softmax(logits)
coder = ImageCoder()
files = get_files(path)
gender_dict = {}
try:
for f in files:
best_choice = classify(sess, label_list, softmax_output, coder, images, f)
# print(best_choice)
gender_dict[f[len(path) + 1:]] = best_choice
return(best_choice)
except Exception as e:
print(e)
print('Failed to run image %s ' % file)
def guessAge(image_file):
#import!!!Fix the bug http://stackoverflow.com/questions/33765336/remove-nodes-from-graph-or-reset-entire-default-graph
tf.reset_default_graph()
with tf.Session() as sess:
age_label_list = AGE_LIST
agelabels = len(age_label_list)
# print('Executing on %s' % FLAGS.device_id)
model_fn = select_model('inception')
images = tf.placeholder(tf.float32,
[None, RESIZE_FINAL, RESIZE_FINAL, 3])
logits_age = model_fn(agelabels, images, 1, False)
init = tf.global_variables_initializer()
requested_step = FLAGS.requested_step if FLAGS.requested_step else None
checkpoint_path = '%s' % (AGE_MODEL_PATH)
# update in 0.11 version
model_checkpoint_path, global_step = get_checkpoint(
checkpoint_path, requested_step, FLAGS.checkpoint)
#print 'model_checkpoint_path is', model_checkpoint_path
#print model_checkpoint_path
saver = tf.train.Saver()
if not saver.last_checkpoints:
saver.restore(sess, model_checkpoint_path)
softmax_output = tf.nn.softmax(logits_age)
coder = ImageCoder()
files = []
# detect age
best_choice = classify(sess, age_label_list, softmax_output, coder, images,
image_file)
sess.close()
return best_choice
def main(argv=None): # pylint: disable=unused-argument
with tf.Session() as sess:
label_list = AGE_LIST if FLAGS.class_type == 'age' else GENDER_LIST
nlabels = len(label_list)
print('Executing on %s' % FLAGS.device_id)
model_fn = select_model(FLAGS.model_type)
images = tf.placeholder(tf.float32,
[None, RESIZE_FINAL, RESIZE_FINAL, 3])
logits = model_fn(nlabels, images, 1, False)
init = tf.initialize_all_variables()
requested_step = FLAGS.requested_step if FLAGS.requested_step else None
checkpoint_path = '%s' % (FLAGS.model_dir)
model_checkpoint_path, global_step = get_checkpoint(
checkpoint_path, requested_step, FLAGS.checkpoint)
saver = tf.train.Saver()
saver.restore(sess, model_checkpoint_path)
softmax_output = tf.nn.softmax(logits)
coder = ImageCoder()
files = []
if FLAGS.face_detection_model:
print('Using face detector %s' % FLAGS.face_detection_model)
face_detect = FaceDetector(FLAGS.face_detection_model)
face_files, rectangles = face_detect.run(FLAGS.filename)
files += face_files
if len(files) == 0:
files.append(FLAGS.filename)
for f in files:
classify(sess, label_list, softmax_output, coder, images, f)
def guessGender(image_file):
tf.reset_default_graph()
with tf.Session() as sess:
#sess = tf.Session()
age_label_list = AGE_LIST
gender_label_list = GENDER_LIST
genderlabels = len(gender_label_list)
# print('Executing on %s' % FLAGS.device_id)
model_fn = select_model('')
with tf.device(FLAGS.device_id):
images = tf.placeholder(tf.float32,
[None, RESIZE_FINAL, RESIZE_FINAL, 3])
logits_gender = model_fn(genderlabels, images, 1, False)
init = tf.global_variables_initializer()
requested_step = FLAGS.requested_step if FLAGS.requested_step else None
checkpoint_path = '%s' % (GENDER_MODEL_PATH)
model_checkpoint_path, global_step = get_checkpoint(
checkpoint_path, requested_step, FLAGS.checkpoint)
saver = tf.train.Saver()
saver.restore(sess, model_checkpoint_path)
softmax_output = tf.nn.softmax(logits_gender)
coder = ImageCoder()
files = []
# detect gender
#try:
best_choice = classifyGender(sess, gender_label_list,
softmax_output, coder, images,
image_file)
return best_choice
def guessGender(image_file):
with tf.Session() as sess:
sess = tf.Session()
age_label_list = AGE_LIST
gender_label_list = GENDER_LIST
genderlabels = len(gender_label_list)
# print('Executing on %s' % FLAGS.device_id)
model_fn = select_model('inception')
images = tf.placeholder(tf.float32,
[None, RESIZE_FINAL, RESIZE_FINAL, 3])
logits_gender = model_fn(genderlabels, images, 1, False)
init = tf.global_variables_initializer()
requested_step = FLAGS.requested_step if FLAGS.requested_step else None
checkpoint_path = '%s' % (FLAGS.model_dir)
model_checkpoint_path, global_step = get_checkpoint(
checkpoint_path, requested_step, FLAGS.checkpoint)
saver = tf.train.Saver()
saver.restore(sess, model_checkpoint_path)
softmax_output = tf.nn.softmax(logits_gender)
coder = ImageCoder()
files = []
# detect age
try:
best_choice = classify(sess, gender_label_list, softmax_output,
coder, images, image_file)
return best_choice
except Exception as e:
print(e)
print('Failed to run image %s ' % image_file)
def model_init(sess, model_path, label_list):
model_checkpoint_path, global_step = get_checkpoint(
model_path, None, FLAGS.checkpoint)
nlabels = len(label_list)
model_fn = select_model(FLAGS.model_type)
images_placeholder = tf.placeholder(tf.float32,
[None, RESIZE_FINAL, RESIZE_FINAL, 3])
with tf.device(FLAGS.device_id):
logits = model_fn(nlabels, images_placeholder, 1, False)
init = tf.global_variables_initializer()
saver = tf.train.Saver()
saver.restore(sess, model_checkpoint_path)
softmax_output = tf.nn.softmax(logits)
return softmax_output, images_placeholder
def __init__(self,session_conf,class_type,model_dir,requested_step='',model_type='inception'):
self.graph=tf.Graph()#为每个类(实例)单独创建一个graph
self.session_conf=session_conf
self.sess=tf.Session(graph=self.graph,config=self.session_conf)#创建新的sess
self.label_list = AGE_LIST if class_type == 'age' else GENDER_LIST
self.model_dir=model_dir
self.requested_step=requested_step
with self.sess.as_default():
with self.graph.as_default():
nlabels = len(self.label_list)
print('Executing on %s' % device_id)
model_fn = select_model(model_type)
with tf.device(device_id):
self.images = tf.placeholder(tf.float32, [None, RESIZE_FINAL, RESIZE_FINAL, 3])
logits = model_fn(nlabels, self.images, 1, False)
init = tf.global_variables_initializer()
requested_step = self.requested_step if self.requested_step else None
checkpoint_path = '%s' % (self.model_dir)
######gender_model_dir
model_checkpoint_path, global_step = get_checkpoint(checkpoint_path, requested_step, checkpoint)
self.saver = tf.train.Saver()
self.saver.restore(self.sess, model_checkpoint_path)#从恢复点恢复参数
self.softmax_output = tf.nn.softmax(logits)
self.coder = ImageCoder()
def main(argv=None):
with tf.Graph().as_default():
model_fn = select_model(FLAGS.model_type)
# Open the metadata file and figure out nlabels, and size of epoch
input_file = os.path.join(FLAGS.train_dir, 'md.json')
print(input_file)
with open(input_file, 'r') as f:
md = json.load(f)
images, labels, _ = distorted_inputs(FLAGS.train_dir, FLAGS.batch_size,
FLAGS.image_size,
FLAGS.num_preprocess_threads)
logits = model_fn(md['nlabels'], images, 1 - FLAGS.pdrop, True)
total_loss = loss(logits, labels)
ini_global_step = 0
train_op = optimizer(FLAGS.optim, FLAGS.eta, total_loss,
FLAGS.steps_per_decay, FLAGS.eta_decay_rate)
saver = tf.train.Saver(tf.global_variables())
summary_op = tf.summary.merge_all()
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=FLAGS.log_device_placement))
tf.global_variables_initializer().run(session=sess)
# This is total hackland, it only works to fine-tune iv3
if FLAGS.pre_model:
inception_variables = tf.get_collection(tf.GraphKeys.VARIABLES,
scope="InceptionV3")
restorer = tf.train.Saver(inception_variables)
restorer.restore(sess, FLAGS.pre_model)
if FLAGS.pre_checkpoint_path:
if tf.gfile.Exists(FLAGS.pre_checkpoint_path) is True:
print('Trying to restore checkpoint from %s' %
FLAGS.pre_checkpoint_path)
restorer = tf.train.Saver()
restorer.restore(
sess,
tf.train.latest_checkpoint(FLAGS.pre_checkpoint_path))
print('%s: Pre-trained model restored from %s' %
(datetime.now(), FLAGS.pre_checkpoint_path))
ini_global_step = get_restored_step(FLAGS.pre_checkpoint_path)
print('Initail Global Step is {}'.format(ini_global_step))
run_dir = '%s/run-%d' % (FLAGS.train_dir, os.getpid())
checkpoint_path = '%s/%s' % (run_dir, FLAGS.checkpoint)
if tf.gfile.Exists(run_dir) is False:
print('Creating %s' % run_dir)
tf.gfile.MakeDirs(run_dir)
tf.train.write_graph(sess.graph_def, run_dir, 'model.pb', as_text=True)
tf.train.start_queue_runners(sess=sess)
summary_writer = tf.summary.FileWriter(run_dir, sess.graph)
steps_per_train_epoch = int(md['train_counts'] / FLAGS.batch_size)
num_steps = FLAGS.max_steps if FLAGS.epochs < 1 else FLAGS.epochs * steps_per_train_epoch
print('Requested number of steps [%d]' % num_steps)
for step in xrange(num_steps):
step += ini_global_step
start_time = time.time()
_, loss_value = sess.run([train_op, total_loss])
duration = time.time() - start_time
assert not np.isnan(loss_value), 'Model diverged with loss = NaN'
if step % 10 == 0:
num_examples_per_step = FLAGS.batch_size
examples_per_sec = num_examples_per_step / duration
sec_per_batch = float(duration)
format_str = (
'%s: step %d, loss = %.3f (%.1f examples/sec; %.3f '
'sec/batch)')
print(format_str % (datetime.now(), step, loss_value,
examples_per_sec, sec_per_batch))
# Loss only actually evaluated every 100 steps?
if step % 100 == 0:
summary_str = sess.run(summary_op)
summary_writer.add_summary(summary_str, step)
if step % 1000 == 0 or (step + 1) == num_steps:
saver.save(sess, checkpoint_path, global_step=step)
def detectUndetectedPersons(outfile, undetected_persons):
#RUDE CARNIE DEFAULTS
print("starting the process to detect people's age and gender...")
gender_model_dir = "./age_and_gender_detection/pretrained_checkpoints/gender/"
age_model_dir = "./age_and_gender_detection/pretrained_checkpoints/age/"
# What processing unit to execute inference on
device_id = '/device:GPU:0'
# Checkpoint basename
checkpoint = 'checkpoint'
model_type = 'inception'
gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=0.1)
config = tf.ConfigProto(allow_soft_placement=True)
# gpu_options=gpu_options)
# config = tf.ConfigProto(allow_soft_placement=True)
with tf.Session(config=config) as sess:
#Age detection model
n_ages = len(AGE_LIST)
age_model_fn = select_model(model_type)
#Gender detection model
n_genders = len(GENDER_LIST)
gender_model_fn = select_model(model_type)
print("initializing the model to detect age and gender")
with tf.device(device_id):
print "initializing the model to detect age and gender using ", str(
device_id)
images = tf.placeholder(tf.float32,
[None, RESIZE_FINAL, RESIZE_FINAL, 3])
requested_step = None
init = tf.global_variables_initializer()
#age model
age_logits = age_model_fn("age", n_ages, images, 1, False)
age_checkpoint_path, global_step = get_checkpoint(
age_model_dir, requested_step, checkpoint)
age_vars = set(tf.global_variables())
saver_age = tf.train.Saver(list(age_vars))
saver_age.restore(sess, age_checkpoint_path)
age_softmax_output = tf.nn.softmax(age_logits)
#gender_model
gender_logits = gender_model_fn("gender", n_genders, images, 1,
False)
gender_checkpoint_path, global_step = get_checkpoint(
gender_model_dir, requested_step, checkpoint)
gender_vars = set(tf.global_variables()) - age_vars
saver_gender = tf.train.Saver(list(gender_vars))
saver_gender.restore(sess, gender_checkpoint_path)
gender_softmax_output = tf.nn.softmax(gender_logits)
coder = ImageCoder()
writer = None
print(
"starting the loop for detecting age and gender in each frame")
time.sleep(
15) # sleep to allow the tensor flow/rude carnie stuff to load
for person_name, person_img in undetected_persons:
outfile.write("%s%s%s"%(person_name, getAgeAndGender(person_name, person_img, sess, coder, images,\
writer, AGE_LIST, GENDER_LIST, age_softmax_output,\
gender_softmax_output), '\n'))
def test(config):
with open(config.word_emb_file, "r") as fh:
word_mat = np.array(json.load(fh), dtype=np.float32)
with open(config.char_emb_file, "r") as fh:
char_mat = np.array(json.load(fh), dtype=np.float32)
with open(config.dev_eval_file, "r") as fh:
eval_file = json.load(fh)
with open(config.dev_example, "r") as fh:
example_file = json.load(fh)
# # with open(config.test_eval_file, "r") as fh:
# # eval_file = json.load(fh)
#
# # dp_test = DataProcessor('test', config)
# # total = dp_test.get_data_size()
dp_test = DataProcessor('test', config, is_test=True)
total = dp_test.get_data_size()
print("Loading model...")
sess_config = tf.ConfigProto(allow_soft_placement=True)
sess_config.gpu_options.allow_growth = True
with tf.Session(config=sess_config) as sess:
selector = select_model(sess, config, word_mat, char_mat)
spaner = span_model(sess, config, word_mat, char_mat)
sess.run(tf.global_variables_initializer())
variables = tf.contrib.framework.get_variables_to_restore()
variables_to_restore = variables
#variables_to_restore = [v for v in variables if v.name.split('/')[0] != 'output']
saver = tf.train.Saver()
# saver = tf.train.Saver()
saver.restore(sess,
tf.train.latest_checkpoint(config.span_save_dir_trained))
'''
from tensorflow.python import pywrap_tensorflow
checkpoint_path = tf.train.latest_checkpoint(config.span_save_dir_trained)
reader = pywrap_tensorflow.NewCheckpointReader(checkpoint_path)
var_to_shape_map = reader.get_variable_to_shape_map()
reader = tf.train.NewCheckpointReader(tf.train.latest_checkpoint(config.span_save_dir_trained))
variables = reader.get_variable_to_shape_map()
saver.restore(sess, tf.train.latest_checkpoint(config.span_save_dir_trained))
'''
sess.run(
tf.assign(selector.is_train, tf.constant(False, dtype=tf.bool)))
sess.run(tf.assign(spaner.is_train, tf.constant(False, dtype=tf.bool)))
losses = []
answer_dict = {}
remapped_dict = {}
for step in tqdm(range(total // config.batch_size)):
tensor_dict, _ = dp_test.get_train_batch(step, is_test=True)
qa_id, yp, outer, para_encode, para_enc_mask, q_emb, sen_emb, lo = sess.run(
[
selector.qa_id, selector.yp, selector.outer,
selector.att_s, selector.c_p_mask, selector.q_emb,
selector.c_emb, selector.lo
],
feed_dict={
selector.qa_id: tensor_dict['ids'],
selector.q: tensor_dict['ques_idxs'],
selector.cs: tensor_dict['context_s_idxs'],
selector.y: tensor_dict['y'],
selector.ce: tensor_dict['context_s_exist_tag'],
selector.ct: tensor_dict['context_type_tag'],
selector.qt: tensor_dict['ques_type_tag']
})
select_sentence = []
sentences_len = []
q = []
for i in range(config.batch_size):
ques = np.zeros([config.ques_limit, q_emb.shape[-1]],
np.float32)
ques[:q_emb.shape[-2]] = q_emb[i]
q.append(ques)
sentences = []
sentence_len = []
sum = tensor_dict['sentence_num'][i]
indexs = np.argsort(-outer[i])
for j in range(config.k):
top_index = indexs[j]
sentence = np.zeros([config.sen_len, sen_emb.shape[-1]],
np.float32)
sentence[:sen_emb.shape[-2]] = sen_emb[i][top_index]
sentences.append(sentence)
sentence_length = np.arange(
sum[indexs[j] - 1],
sum[indexs[j] - 1] + config.sen_len, 1)
sentence_len.append(sentence_length)
sentence_len = np.array(sentence_len)
sentences = np.array(sentences)
select_sentence.append(sentences)
sentences_len.append(sentence_len)
select_sentences = np.array(select_sentence)
sentences_lens = np.array(sentences_len)
q = np.array(q)
qa_id, loss, train_op_span, yp1, yp2 = sess.run(
[
spaner.qa_id, spaner.loss_span, spaner.train_op,
spaner.yp1, spaner.yp2
],
feed_dict={
spaner.qa_id: tensor_dict['ids'],
spaner.para: tensor_dict['para_idxs'],
spaner.para_e: tensor_dict['para_exist_tag'],
spaner.para_t: tensor_dict['para_type_tag'],
spaner.q: q,
spaner.sentence: select_sentences,
spaner.sentence_index: sentences_lens,
spaner.outer: outer,
spaner.para_enc: para_encode,
spaner.para_enc_mask: para_enc_mask,
spaner.y1: tensor_dict['y1'],
spaner.y2: tensor_dict['y2']
})
answer_dict_, remapped_dict_ = convert_tokens(
eval_file, qa_id.tolist(), yp1.tolist(), yp2.tolist())
answer_dict.update(answer_dict_)
remapped_dict.update(remapped_dict_)
losses.append(loss)
loss = np.mean(losses)
metrics = evaluate(eval_file, answer_dict)
with open(config.answer_file, "w") as fh:
json.dump(remapped_dict, fh)
print("Exact Match: {}, F1: {}".format(metrics['exact_match'],
metrics['f1']))
def main(argv=None):
files = []
#if FLAGS.facedetect_model_type is 'MTCNN':
# print('Using face detector %s' %(FLAGS.facedetect_model_type))
# img = cv2.imread(FLAGS.filename)
# img_resize = cv2.resize(img, (640,480))
# img_resize = cv2.cvtColor(img_resize, cv2.COLOR_BGR2RGB)
# faceimagesequence, faceimagelocation, faceimagelandmarks, numberoffaces = MTCNNDetectFace(image)
# Load Model
if FLAGS.multitask:
config = tf.ConfigProto(allow_soft_placement=True)
with tf.Session(config=config) as sess:
age_nlabels = len(AGE_LIST)
gender_nlabels = len(GENDER_LIST)
print('Executing on %s ' % FLAGS.device_id)
model_fn = select_model(FLAGS.model_type)
with tf.device(FLAGS.device_id):
images = tf.placeholder(tf.float32, [None,RESIZE_FINAL,RESIZE_FINAL, 3], name='input')
agelogits, genderlogits = model_fn(age_nlabels,images, gender_nlabels, images, 1, False)
init = tf.global_variables_initializer()
requested_step = FLAGS.requested_step if FLAGS.requested_step else None
checkpoint_path = '%s' % (FLAGS.model_dir)
model_checkpoint_path, global_step = get_checkpoint(checkpoint_path, requested_step, FLAGS.checkpoint)
saver = tf.train.Saver()
saver.restore(sess, model_checkpoint_path)
softmax_age_output = tf.nn.softmax(agelogits, name='ageoutput')
softmax_gender_output = tf.nn.softmax(genderlogits, name='genderoutput')
print (softmax_age_output)
print (softmax_gender_output)
coder = ImageCoder()
#support a batch mode if no face detection model
if len(files) == 0:
if (os.path.isdir(FLAGS.filename)):
for relpath in os.listdir(FLAGS.filename):
abspath = os.path.join(FLAGS.filename, relpath)
if os.path.isfile(abspath) and any([abspath.endswith('.'+ty) for ty in ('jpg', 'png', 'JPG', 'PNG', 'jpeg')]):
print(abspath)
files.append(abspath)
else:
files.append(FLAGS.filename)
# if it happens to be a list file, read the list and clobber the files
if any([FLAGS.filename.endswith('.' + ty) for ty in ('csv', 'tsv', 'txt')]):
files = list_images(FLAGS.filename)
writer = None
output = None
if FLAGS.resultfile:
print('Creating output file %s ' % FLAGS.resultfile)
output = open(FLAGS.resultfile, 'w')
writer = csv.writer(output)
writer.writerow(('file', 'label', 'score'))
image_files = list(filter(lambda x: x is not None, [resolve_file(f) for f in files]))
print(image_files)
if FLAGS.single_look:
classify_many_single_crop(sess, AGE_LIST, softmax_age_output,
coder, images, image_files, writer)
classify_many_single_crop(sess, GENDER_LIST, softmax_gender_output,
coder, images, image_files, writer)
else:
for image_file in image_files:
classify_one_multi_crop(sess, AGE_LIST, softmax_age_output,
coder, images, image_file, writer)
classify_one_multi_crop(sess, GENDER_LIST, softmax_gender_output,
coder, images, image_file, writer)
if output is not None:
output.close()
if FLAGS.convertpb:
# retrieve the protobuf graph definition
graph = tf.get_default_graph()
input_graph_def = graph.as_graph_def()
output_node_names = 'ageoutput_1,genderoutput_1'
output_graph_def = graph_util.convert_variables_to_constants(
sess, #The session is used to retrieve the weights
input_graph_def, #The graph_def is used to retrieve the nodes
output_node_names.split(",") #The output node names are used to select the usefull nodes
)
# finally we serialize and dump the output graph to the filesystem
output_pb_file = FLAGS.model_dir+os.sep+FLAGS.model_type+'.pb'
with tf.gfile.GFile(output_pb_file, "wb") as f:
f.write(output_graph_def.SerializeToString())
print("%d ops in the final graph." % len(output_graph_def.node) )
else:
config = tf.ConfigProto(allow_soft_placement=True)
with tf.Session(config=config) as sess:
if FLAGS.class_type == 'Age':
label_list = AGE_LIST
elif FLAGS.class_type == 'Gender':
label_list = GENDER_LIST
nlabels = len(label_list)
print('Executing on %s' % FLAGS.device_id)
model_fn = select_model(FLAGS.model_type)
with tf.device(FLAGS.device_id):
images = tf.placeholder(tf.float32, [None, RESIZE_FINAL, RESIZE_FINAL, 3],name='input')
logits = model_fn(nlabels, images, 1, False)
init = tf.global_variables_initializer()
requested_step = FLAGS.requested_step if FLAGS.requested_step else None
checkpoint_path = '%s' % (FLAGS.model_dir)
model_checkpoint_path, global_step = get_checkpoint(checkpoint_path, requested_step, FLAGS.checkpoint)
saver = tf.train.Saver()
saver.restore(sess, model_checkpoint_path)
if FLAGS.class_type == 'Age':
softmax_output = tf.nn.softmax(logits, name='ageoutput')
elif FLAGS.class_type == 'Gender':
softmax_output = tf.nn.softmax(logits, name='genderoutput')
coder = ImageCoder()
# Support a batch mode if no face detection model
if len(files) == 0:
if (os.path.isdir(FLAGS.filename)):
for relpath in os.listdir(FLAGS.filename):
abspath = os.path.join(FLAGS.filename, relpath)
if os.path.isfile(abspath) and any([abspath.endswith('.' + ty) for ty in ('jpg', 'png', 'JPG', 'PNG', 'jpeg')]):
print(abspath)
files.append(abspath)
else:
files.append(FLAGS.filename)
# If it happens to be a list file, read the list and clobber the files
if any([FLAGS.filename.endswith('.' + ty) for ty in ('csv', 'tsv', 'txt')]):
files = list_images(FLAGS.filename)
writer = None
output = None
if FLAGS.resultfile:
print('Creating output file %s ' % FLAGS.resultfile)
output = open(FLAGS.resultfile, 'w')
writer = csv.writer(output)
writer.writerow(('file', 'label', 'score'))
image_files = list(filter(lambda x: x is not None, [resolve_file(f) for f in files]))
print(image_files)
if FLAGS.single_look:
classify_many_single_crop(sess, label_list, softmax_output, coder, images, image_files, writer)
else:
for image_file in image_files:
classify_one_multi_crop(sess, label_list, softmax_output, coder, images, image_file, writer)
if output is not None:
output.close()
if FLAGS.convertpb:
# retrieve the protobuf graph definition
graph = tf.get_default_graph()
input_graph_def = graph.as_graph_def()
if FLAGS.class_type == 'Age':
output_node_names = 'ageoutput'
elif FLAGS.class_type == 'Gender':
output_node_names = 'genderoutput'
output_graph_def = graph_util.convert_variables_to_constants(
sess, #The session is used to retrieve the weights
input_graph_def, #The graph_def is used to retrieve the nodes
output_node_names.split(",") #The output node names are used to select the usefull nodes
)
# finally we serialize and dump the output graph to the filesystem
output_pb_file = FLAGS.model_dir+os.sep+FLAGS.model_type+'_'+FLAGS.class_type+'.pb'
with tf.gfile.GFile(output_pb_file, "wb") as f:
f.write(output_graph_def.SerializeToString())
print("%d ops in the final graph." % len(output_graph_def.node) )
def guess(model_dir='./21936', class_type='genda', model_type='inception',filename='', device_id='/cpu:0', requested_step='', target='future_genda_prediction', checkpoint='14999', face_detection_model='', face_detection_type='cascade',count=''):
files = []
if face_detection_model:
print('Using face detector (%s) %s' % (face_detection_type, face_detection_model))
face_detect = face_detection_model(face_detection_type, face_detection_model)
face_files, rectangles = face_detect.run(filename)
print(face_files)
files += face_files
config = tf.ConfigProto(allow_soft_placement=True)
with tf.Session(config=config) as sess:
if class_type == 'genda':
label_list = GENDER_LIST
elif class_type == 'age':
label_list = AGE_LIST
nlabels = len(label_list)
model_fn = select_model(model_type)
with tf.device(device_id):
images = tf.placeholder(tf.float32, [None, RESIZE_FINAL, RESIZE_FINAL, 3])
logits = model_fn(nlabels, images, 1, False,count)
init = tf.global_variables_initializer()
requested_step = requested_step if requested_step else None
checkpoint_path = '%s' % (model_dir)
model_checkpoint_path, global_step = get_checkpoint(checkpoint_path, requested_step, checkpoint)
#加载tf模型的方法
if count==0:
startT=time.time()
saver = tf.train.Saver()
saver.restore(sess, model_checkpoint_path)
endT = time.time()
loadms = (endT - startT) * 1000
print(loadms)
softmax_output = tf.nn.softmax(logits)
coder = ImageCoder()
# Support a batch mode if no face detection model
if len(files) == 0:
if (os.path.isdir(filename)):
for relpath in os.listdir(filename):
abspath = os.path.join(filename, relpath)
if os.path.isfile(abspath) and any([abspath.endswith('.' + ty) for ty in ('jpg', 'png', 'JPG', 'PNG', 'jpeg')]):
print(abspath)
files.append(abspath)
else:
files.append(filename)
# If it happens to be a list file, read the list and clobber the files
if any([filename.endswith('.' + ty) for ty in ('csv', 'tsv', 'txt')]):
files = list_images(filename)
writer = None
output = None
if target:
print('Creating output file %s' % target)
output = open(target, 'w')
writer = csv.writer(output)
writer.writerow(('file', 'label', 'score'))
image_files = list(filter(lambda x: x is not None, [resolve_file(f) for f in files]))
print(image_files)
for image_file in image_files: #需要将files中的多个结果综合一下
classify_one_multi_crop(sess, init,label_list, softmax_output, coder, images, image_file, writer)
if output is not None:
output.close()
def train(config):
with open(config.word_emb_file, "r") as fh:
word_mat = np.array(json.load(fh), dtype=np.float32)
with open(config.char_emb_file, "r") as fh:
char_mat = np.array(json.load(fh), dtype=np.float32)
with open(config.train_eval_file, "r") as fh:
train_eval_file = json.load(fh)
with open(config.dev_eval_file, "r") as fh:
dev_eval_file = json.load(fh)
dp_train = DataProcessor('train', config)
dp_dev = DataProcessor('dev', config)
#dp_dev = dp_train
dev_total = dp_dev.get_data_size()
print("Building model...")
sess_config = tf.ConfigProto(allow_soft_placement=True)
sess_config.gpu_options.allow_growth = True
loss_save = 100.0
patience = 0
lr = config.init_lr
lr_span = config.init_lr_span
train_batch_num = int(np.floor(
dp_train.num_samples / config.batch_size)) - 1
batch_no = 0
SN = config.k
with tf.Session(config=sess_config) as sess:
selector = select_model(sess, config, word_mat, char_mat)
spaner = span_model(sess, config, word_mat, char_mat)
writer = tf.summary.FileWriter(config.RL_log_dir)
sess.run(tf.global_variables_initializer())
#variables = tf.contrib.framework.get_variables_to_restore()
#variables_to_restore = variables
# #variables_to_restore = [v for v in variables if v.name.split('/')[0] != 'output']
# #
#saver_selector = tf.train.Saver(variables_to_restore)
# # saver_selector.restore(sess, tf.train.latest_checkpoint(config.save_dir))
saver = tf.train.Saver()
saver.restore(sess,
tf.train.latest_checkpoint(config.span_save_dir_trained))
sess.run(tf.assign(selector.is_train, tf.constant(True,
dtype=tf.bool)))
sess.run(tf.assign(spaner.is_train, tf.constant(True, dtype=tf.bool)))
sess.run(tf.assign(selector.lr, tf.constant(lr, dtype=tf.float32)))
sess.run(tf.assign(spaner.lr, tf.constant(lr_span, dtype=tf.float32)))
sen_len_limit = config.sen_len
question_type_num = 19
for batch_time in tqdm(range(1, config.num_steps + 1)):
tensor_dict, _ = dp_train.get_train_batch(batch_no %
train_batch_num)
sess.run(
tf.assign(selector.is_train, tf.constant(False,
dtype=tf.bool)))
sess.run(
tf.assign(spaner.is_train, tf.constant(True, dtype=tf.bool)))
qa_id, yp, outer, para_encode, para_enc_mask, q_emb, sen_emb, lo = sess.run(
[
selector.qa_id, selector.yp, selector.outer,
selector.att_s, selector.c_p_mask, selector.q_emb,
selector.c_emb, selector.lo
],
feed_dict={
selector.qa_id: tensor_dict['ids'],
selector.q: tensor_dict['ques_idxs'],
selector.cs: tensor_dict['context_s_idxs'],
selector.y: tensor_dict['y'],
selector.ce: tensor_dict['context_s_exist_tag'],
selector.ct: tensor_dict['context_type_tag'],
selector.qt: tensor_dict['ques_type_tag']
})
np.savetxt("lo.txt", lo)
select_sentence = []
sentences_len = []
q = []
sentences_cs = []
sentences_ce = []
sentences_ct = []
for i in range(config.batch_size):
ques = np.zeros([config.ques_limit, q_emb.shape[-1]],
np.float32)
ques[:q_emb.shape[-2]] = q_emb[i]
q.append(ques)
sentences = []
sentence_len = []
sum = tensor_dict['sentence_num'][i]
indexs = np.argsort(-outer[i])
#RL change
# indexs = np.random.choice(a=outer.shape[0], size=config.k, replace=False, p=outer)
for j in range(config.k):
top_index = indexs[j]
sentence = np.zeros([config.sen_len, sen_emb.shape[-1]],
np.float32)
sentence[:sen_emb.shape[-2]] = sen_emb[i][top_index]
sentences.append(sentence)
sentence_length = np.arange(
sum[indexs[j] - 1],
sum[indexs[j] - 1] + config.sen_len, 1)
sentence_len.append(sentence_length)
sentence_len = np.array(sentence_len)
sentences = np.array(sentences)
select_sentence.append(sentences)
sentences_len.append(sentence_len)
select_sentences = np.array(select_sentence)
sentences_lens = np.array(sentences_len)
q = np.array(q)
global_step = sess.run(spaner.global_step) + 1
qa_id, loss, train_op_span, yp1, yp2, topk,ou1,out2,\
loss1,loss2,\
lo1,lo2,lo3,lo4 = sess.run([spaner.qa_id, spaner.loss_span, spaner.train_op,spaner.yp1, spaner.yp2, spaner.topk,
spaner.out_lo1,spaner.out_lo2,
spaner.out1, spaner.out2,
spaner.lo1,spaner.lo2,spaner.lo3,spaner.lo4],
feed_dict={spaner.qa_id: tensor_dict['ids'],
spaner.para: tensor_dict['para_idxs'],
spaner.para_e: tensor_dict['para_exist_tag'],
spaner.para_t: tensor_dict['para_type_tag'],
spaner.q: q,
spaner.sentence: select_sentences,
spaner.sentence_index: sentences_lens,
spaner.outer: outer,
spaner.para_enc: para_encode,
spaner.para_enc_mask: para_enc_mask,
spaner.y1: tensor_dict['y1'],
spaner.y2: tensor_dict['y2']
})
answer_dict, _ = convert_tokens(train_eval_file, qa_id.tolist(),
yp1.tolist(), yp2.tolist())
reward = compute_reward(train_eval_file, answer_dict)
sess.run(
tf.assign(selector.is_train, tf.constant(True, dtype=tf.bool)))
sess.run(
tf.assign(spaner.is_train, tf.constant(False, dtype=tf.bool)))
qa_id, yp, outer, sel_loss, train_op, policy = sess.run(
[
selector.qa_id, selector.yp, selector.outer, selector.loss,
selector.train_op, selector.policy_log_part
],
feed_dict={
selector.qa_id: tensor_dict['ids'],
selector.q: tensor_dict['ques_idxs'],
selector.cs: tensor_dict['context_s_idxs'],
selector.y: tensor_dict['y'],
selector.ce: tensor_dict['context_s_exist_tag'],
selector.ct: tensor_dict['context_type_tag'],
selector.qt: tensor_dict['ques_type_tag'],
selector.reward: reward
})
np.savetxt("topk.txt", topk)
# np.savetxt("out1.txt",ou1)
np.savetxt("l1.txt", lo1)
np.savetxt("l2.txt", lo2)
np.savetxt("l3.txt", lo3)
np.savetxt("l4.txt", lo4)
np.savetxt("lss1.txt", loss1)
np.savetxt("lss2.txt", loss2)
np.savetxt("reward.txt", reward)
np.savetxt("policy.txt", policy)
# np.savetxt("loss3.txt", loss3)
# numpy.savetxt('new.csv', my_matrix, delimiter=',')
#print(loss3)
print(loss)
print("selector_loss" + str(sel_loss))
batch_no = batch_no + 1
if global_step % config.period == 0:
loss_sum = tf.Summary(value=[
tf.Summary.Value(tag="model/loss", simple_value=loss),
])
writer.add_summary(loss_sum, global_step)
if batch_time % config.checkpoint == 0:
sess.run(
tf.assign(selector.is_train,
tf.constant(False, dtype=tf.bool)))
sess.run(
tf.assign(spaner.is_train, tf.constant(False,
dtype=tf.bool)))
# _, summ = evaluate_batch(
# selector, config.val_num_batches, train_eval_file, sess, "train", dp_train)
metrics, summ = evaluate_span_batch(
config, selector, spaner, dev_total // config.batch_size,
dev_eval_file, sess, "dev", dp_dev)
for s in summ:
writer.add_summary(s, global_step)
print('epoch' + str(global_step))
print('dev_loss:' + str(metrics["loss"]))
print('F1:' + str(metrics["f1"]))
print('em:' + str(metrics["exact_match"]))
sess.run(
tf.assign(spaner.is_train, tf.constant(True,
dtype=tf.bool)))
dev_loss = metrics["loss"]
if dev_loss < loss_save:
loss_save = dev_loss
patience = 0
else:
patience += 1
if patience >= config.patience:
lr /= 2.0
loss_save = dev_loss
patience = 0
sess.run(
tf.assign(spaner.lr, tf.constant(lr, dtype=tf.float32)))
for s in summ:
writer.add_summary(s, global_step)
writer.flush()
filename = os.path.join(config.RL_save_dir,
"model_{}.ckpt".format(global_step))
saver.save(sess, filename)
def construct(filename,
class_type,
model_type,
model_dir,
checkpoint='checkpoint',
device='/cpu:0',
target=None,
classes=None): # pylint: disable=unused-argument
# sys.stdout = os.devnull
# sys.stderr = os.devnull
files = []
with tf.Graph().as_default():
with tf.Session() as sess:
#with tf.Session() as sess:
#print('\n111111111111111\n')
#tf.reset_default_graph()
label_list = AGE_LIST if class_type == 'age' else GENDER_LIST
nlabels = len(label_list)
#print('\n222222222222222\n')
#print('Executing on %s' % FLAGS.device_id)
model_fn = select_model(model_type)
with tf.device(device):
# sys.stdout = sys.__stdout__
# sys.stderr = sys.__stderr__
images = tf.placeholder(tf.float32,
[None, RESIZE_FINAL, RESIZE_FINAL, 3])
logits = model_fn(nlabels, images, 1, False)
init = tf.global_variables_initializer()
#print('\n333333333333333\n')
requested_step = None
checkpoint_path = '%s' % (model_dir)
model_checkpoint_path, global_step = get_checkpoint(
checkpoint_path, requested_step, checkpoint)
#print("\nglobal_step=",global_step)
saver = tf.train.Saver()
#print('\n44444444444444444\n')
#print("PATH=",model_checkpoint_path,'\n')
saver.restore(sess, model_checkpoint_path)
#print('\n55555555555555555\n')
softmax_output = tf.nn.softmax(logits)
coder = ImageCoder()
# Support a batch mode if no face detection model
if len(files) == 0:
files.append(filename)
# If it happens to be a list file, read the list and clobber the files
if one_of(filename, ('csv', 'tsv', 'txt')):
files = batchlist(filename)
for it, f in enumerate(files):
image_file = resolve_file(f)
if image_file is None: continue
try:
best_choice = classify(sess, label_list,
softmax_output, coder, images,
image_file)
#results[it][0]=f
#print('f=%s\nresult='%f,results)
print("\nClass_type=", class_type)
if class_type == 'age':
#print("\n%s\n"%it)
classes[it].age = best_choice[0]
# print(best_choice[0],'\n')
# print(results,'\n')
target.writerow(
(f, classes[it].gender, classes[it].age,
'%.2f' % best_choice[1]))
if class_type == 'gender':
#print("\n222222222\n")
classes[it].name = f
classes[it].gender = best_choice[0]
except Exception as e:
print(e)
print('Failed to run image %s ' % image_file)
it += 1
#if output is not None:
# output.close()
# print(results)
sess.close()
from model import select_model, get_checkpoint
from tensorflow.python.framework.graph_util import convert_variables_to_constants
with tf.Session(config=tf.ConfigProto(
allow_soft_placement=True)) as sess:
# tf.reset_default_graph()
model_type = 'inception'
if False: #gender
nlabels = 2
model_dir = '21936'
outname = 'gender_net'
else:
nlabels = 8
model_dir = '22801'
outname = 'age_net'
model_fn = select_model('inception')
with tf.device('/gpu:0'):
images = tf.placeholder(tf.float32,
[None, RESIZE_FINAL, RESIZE_FINAL, 3])
logits = model_fn(nlabels, images, 1, False)
init = tf.global_variables_initializer()
requested_step = None
checkpoint_path = '%s' % (os.path.join(src_dir, 'inception',
model_dir))
model_checkpoint_path, global_step = get_checkpoint(
checkpoint_path, requested_step, 'checkpoint')
saver = tf.train.Saver()
saver.restore(sess, model_checkpoint_path)
softmax_output = tf.nn.softmax(logits, name='logits')
def main(argv=None):
if not os.path.exists(FLAGS.model_dir):
os.mkdir(FLAGS.model_dir)
folddirlist = FLAGS.train_dir.split(os.sep)
subdir = FLAGS.model_dir + os.sep + folddirlist[-2]
if not os.path.exists(subdir):
os.mkdir(subdir)
savemodeldir = subdir + os.sep + folddirlist[-1]
if not os.path.exists(savemodeldir):
os.mkdir(savemodeldir)
if FLAGS.multitask:
with tf.Graph().as_default():
model_fn = select_model(FLAGS.model_type)
# Open the metadata file and figure out nlabels, and size of epoch
input_file_age = os.path.join(FLAGS.train_dir, 'mdage.json')
input_file_gender = os.path.join(FLAGS.train_dir, 'mdgender.json')
with open(input_file_age, 'r') as fage:
mdage = json.load(fage)
with open(input_file_gender, 'r') as fgender:
mdgender = json.load(fgender)
images, agelabels, genderlabels, _ = multiinputs(
FLAGS.train_dir,
batch_size=FLAGS.batch_size,
image_size=FLAGS.image_size,
train=True,
num_preprocess_threads=FLAGS.num_preprocess_threads,
datatype='train')
agelogits, genderlogits = model_fn(mdage['nlabels'], images,
mdgender['nlabels'], images,
1 - FLAGS.pdrop, True)
agelosses, genderlosses, totallosses = multiloss(
agelogits, agelabels, genderlogits, genderlabels)
agegendertrain_op = optimizer(FLAGS.optim, FLAGS.eta, totallosses,
FLAGS.steps_per_decay,
FLAGS.eta_decay_rate)
saver = tf.train.Saver(tf.global_variables())
summary_op = tf.summary.merge_all()
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=FLAGS.log_device_placement))
tf.global_variables_initializer().run(session=sess)
# fine-tune dp_multitask and mobilenet_multitask
if FLAGS.pre_checkpoint_path:
print('Trying to restore checkpoint from %s ' %
FLAGS.pre_checkpoint_path)
if FLAGS.model_type is 'LMTCNN':
all_variables = tf.get_collection(tf.GraphKeys.VARIABLES,
scope="multitaskdpcnn")
elif FLAGS.model_type is 'mobilenet_multitask':
all_variables = tf.get_collection(
tf.GraphKeys.VARIABLES, scope="MobileNetmultitask")
age_variables = tf.get_collection(tf.GraphKeys.VARIABLES,
scope="ageoutput")
gender_variables = tf.get_collection(tf.GraphKeys.VARIABLES,
scope="genderoutput")
all_variables.extend(age_variables)
all_variables.extend(gender_variables)
restorer = tf.train.Saver(all_variables)
restorer.restore(sess, FLAGS.pre_checkpoint_path)
print('%s: Pre-trained model restored from %s' %
(datetime.now(), FLAGS.pre_checkpoint_path))
run_dir = '%s/%s-run-%d' % (savemodeldir, FLAGS.model_type,
os.getpid())
checkpoint_path = '%s/%s' % (run_dir, FLAGS.checkpoint)
if tf.gfile.Exists(run_dir) is False:
print('Creating %s' % run_dir)
tf.gfile.MakeDirs(run_dir)
tf.train.write_graph(sess.graph_def,
run_dir,
'agegendermodel.pb',
as_text=True)
tf.train.start_queue_runners(sess=sess)
summary_writer = tf.summary.FileWriter(run_dir, sess.graph)
stemps_per_train_epoch = int(mdage['train_counts'] /
FLAGS.batch_size)
num_steps = FLAGS.max_steps if FLAGS.epochs < 1 else FLAGS.epochs * steps_per_train_epoch
print('Requested number of steps [%d]' % num_steps)
for step in xrange(num_steps):
start_time = time.time()
_, totallossvalue, agelossvalue, genderlossvalue, imagesvalue, agelabelsvalue, genderlabelsvalue = sess.run(
[
agegendertrain_op, totallosses, agelosses,
genderlosses, images, agelabels, genderlabels
])
duration = time.time() - start_time
assert not np.isnan(
agelossvalue), 'Model diverged with ageloss = NaN'
assert not np.isnan(
genderlossvalue), 'Model diverged with genderloss = NaN'
assert not np.isnan(
totallossvalue), 'Model diverged with totallossvalue= NaN'
if step % 10 == 0:
num_examples_per_step = FLAGS.batch_size
examples_per_sec = num_examples_per_step / duration
sec_per_batch = float(duration)
format_str = (
'%s: step %d, ageloss= %.3f, genderloss= %.3f , totalloss= %.3f (%.1f examples/sec; %.3f '
'sec/batch)')
print(format_str % (datetime.now(), step, agelossvalue[0],
genderlossvalue[0], totallossvalue,
examples_per_sec, sec_per_batch))
# loss evaluated every 100 steps
if step % 100 == 0:
summary_str = sess.run(summary_op)
summary_writer.add_summary(summary_str, step)
if step % 1000 == 0 or (step + 1) == num_steps:
saver.save(sess, checkpoint_path, global_step=step)
else:
with tf.Graph().as_default():
model_fn = select_model(FLAGS.model_type)
if FLAGS.class_type == 'Age':
input_file_age = os.path.join(FLAGS.train_dir, 'mdage.json')
with open(input_file_age, 'r') as fage:
mdage = json.load(fage)
images, labels, _ = inputs_mod(
FLAGS.train_dir,
batch_size=FLAGS.batch_size,
image_size=FLAGS.image_size,
train=True,
num_preprocess_threads=FLAGS.num_preprocess_threads,
classtype=FLAGS.class_type,
datatype='train')
logits = model_fn(mdage['nlabels'], images, 1 - FLAGS.pdrop,
True)
total_loss = loss(logits, labels)
elif FLAGS.class_type == 'Gender':
input_file_gender = os.path.join(FLAGS.train_dir,
'mdgender.json')
with open(input_file_gender, 'r') as fgender:
mdgender = json.load(fgender)
images, labels, _ = inputs_mod(
FLAGS.train_dir,
batch_size=FLAGS.batch_size,
image_size=FLAGS.image_size,
train=True,
num_preprocess_threads=FLAGS.num_preprocess_threads,
classtype=FLAGS.class_type,
datatype='train')
logits = model_fn(mdgender['nlabels'], images, 1 - FLAGS.pdrop,
True)
total_loss = loss(logits, labels)
train_op = optimizer(FLAGS.optim, FLAGS.eta, total_loss,
FLAGS.steps_per_decay, FLAGS.eta_decay_rate)
saver = tf.train.Saver(tf.global_variables())
summary_op = tf.summary.merge_all()
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=FLAGS.log_device_placement))
tf.global_variables_initializer().run(session=sess)
# it only works to fine-tune inception v3
if FLAGS.pre_model:
inception_variables = tf.get_collection(tf.GraphKeys.VARIABLES,
scope="InceptionV3")
restorer = tf.train.Saver(inception_variables)
restorer.restore(sess, FLAGS.pre_model)
if FLAGS.pre_checkpoint_path:
if tf.gfile.Exists(FLAGS.pre_checkpoint_path) is True:
print('Trying to restore checkpoint from %s' %
FLAGS.pre_checkpoint_path)
restorer = tf.train.Saver()
tf.train.latest_checkpoint(FLAGS.pre_checkpoint_path)
print('%s: Pre-trained model restored from %s' %
(datetime.now(), FLAGS.pre_checkpoint_path))
run_dir = '%s/%s-%s-run-%d' % (savemodeldir, FLAGS.model_type,
FLAGS.class_type, os.getpid())
checkpoint_path = '%s/%s' % (run_dir, FLAGS.checkpoint)
if tf.gfile.Exists(run_dir) is False:
print('Creating %s' % run_dir)
tf.gfile.MakeDirs(run_dir)
tf.train.write_graph(sess.graph_def,
run_dir,
'model.pb',
as_text=True)
tf.train.start_queue_runners(sess=sess)
summary_writer = tf.summary.FileWriter(run_dir, sess.graph)
if FLAGS.class_type == 'Age':
steps_per_train_epoch = int(mdage['train_counts'] /
FLAGS.batch_size)
elif FLAGS.class_type == 'Gender':
steps_per_train_epoch = int(mdgender['train_counts'] /
FLAGS.batch_size)
num_steps = FLAGS.max_steps if FLAGS.epochs < 1 else FLAGS.epochs * steps_per_train_epoch
print('Requested number of steps [%d]' % num_steps)
for step in xrange(num_steps):
start_time = time.time()
_, loss_value = sess.run([train_op, total_loss])
duration = time.time() - start_time
assert not np.isnan(
loss_value), 'Model diverged with loss = NaN'
if step % 10 == 0:
num_examples_per_step = FLAGS.batch_size
examples_per_sec = num_examples_per_step / duration
sec_per_batch = float(duration)
format_str = (
'%s: step %d, loss = %.3f (%.1f examples/sec; %.3f '
'sec/batch)')
print(format_str % (datetime.now(), step, loss_value,
examples_per_sec, sec_per_batch))
# loss evaluated every 100 steps
if step % 100 == 0:
summary_str = sess.run(summary_op)
summary_writer.add_summary(summary_str, step)
if step % 1000 == 0 or (step + 1) == num_steps:
saver.save(sess, checkpoint_path, global_step=step)
def main(argv=None):
with tf.Graph().as_default():
global_step = tf.Variable(0, trainable=False)
model_fn = select_model(FLAGS.model_type)
# Open the metadata file and figure out nlabels, and size of epoch
input_file = os.path.join(FLAGS.train_dir, 'md.json')
print(input_file)
with open(input_file, 'r') as f:
md = json.load(f)
images, labels, _ = distorted_inputs(FLAGS.train_dir, FLAGS.batch_size,
FLAGS.image_size,
FLAGS.num_preprocess_threads)
if not FLAGS.dual:
logits = model_fn(md['nlabels'], images, 1 - FLAGS.pdrop, True)
total_loss, accuracy = loss(logits, labels, global_step)
else:
with tf.variable_scope("net1") as scope:
logits1 = model_fn(md['nlabels'], images, 1 - FLAGS.pdrop,
True)
with tf.variable_scope("net2") as scope:
logits2 = model_fn(md['nlabels'], images, 1 - FLAGS.pdrop,
True)
pred1 = tf.argmax(logits1, 1)
pred2 = tf.argmax(logits2, 1)
update_step = tf.stop_gradient(
tf.to_float(
tf.logical_or(tf.not_equal(pred1, pred2),
global_step < FLAGS.init_iter)))
with tf.variable_scope("net1") as scope:
if FLAGS.min_batch_size == -1:
total_loss1, accuracy1 = loss(logits1, labels, global_step,
None, scope.name)
else:
total_loss1, accuracy1 = loss(logits1, labels, global_step,
update_step, scope.name)
with tf.variable_scope("net2") as scope:
if FLAGS.min_batch_size == -1:
total_loss2, accuracy2 = loss(logits2, labels, global_step,
None, scope.name)
else:
total_loss2, accuracy2 = loss(logits2, labels, global_step,
update_step, scope.name)
disagree_rate = tf.reduce_mean(
tf.to_float(tf.not_equal(pred1, pred2)))
if not FLAGS.dual:
train_op = optimizer(FLAGS.optim, FLAGS.eta, total_loss)
else:
with tf.variable_scope("net1") as scope:
var_net1 = [
var for var in tf.all_variables()
if var.name.startswith("net1")
]
train_op1 = optimizer(FLAGS.optim,
FLAGS.eta,
total_loss1,
variables=var_net1,
name=scope.name)
with tf.variable_scope("net2") as scope:
var_net2 = [
var for var in tf.all_variables()
if var.name.startswith("net2")
]
train_op2 = optimizer(FLAGS.optim,
FLAGS.eta,
total_loss2,
variables=var_net2,
name=scope.name)
saver = tf.train.Saver(tf.all_variables(), max_to_keep=151)
summary_op = tf.merge_all_summaries()
init = tf.initialize_all_variables()
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=FLAGS.log_device_placement))
sess.run(init)
# This is total hackland, it only works to fine-tune iv3
if FLAGS.pre_model:
inception_variables = tf.get_collection(tf.GraphKeys.VARIABLES,
scope="InceptionV3")
restorer = tf.train.Saver(inception_variables)
restorer.restore(sess, FLAGS.pre_model)
if FLAGS.pre_checkpoint_path:
if tf.gfile.Exists(FLAGS.pre_checkpoint_path) is True:
print('Trying to restore checkpoint from %s' %
FLAGS.pre_checkpoint_point)
restorer = tf.train.Saver()
tf.train.latest_checkpoint(FLAGS.pre_checkpoint_path)
print('%s: Pre-trained model restored from %s' %
(datetime.now(), FLAGS.pre_checkpoint_path))
run_dir = '%s/run-%d' % (FLAGS.train_dir, os.getpid())
checkpoint_path = '%s/%s' % (run_dir, FLAGS.checkpoint)
if tf.gfile.Exists(run_dir) is False:
print('Creating %s' % run_dir)
tf.gfile.MakeDirs(run_dir)
tf.train.write_graph(sess.graph_def, run_dir, 'model.pb', as_text=True)
tf.train.start_queue_runners(sess=sess)
summary_writer = tf.train.SummaryWriter(run_dir, sess.graph)
steps_per_train_epoch = int(md['train_counts'] / FLAGS.batch_size)
num_steps = FLAGS.max_steps if FLAGS.epochs < 1 else FLAGS.epochs * steps_per_train_epoch
print('Requested number of steps [%d]' % num_steps)
trainable_buffer_img = None
trainable_buffer_lbl = None
for step in range(num_steps):
start_time = time.time()
if FLAGS.Qloss:
_, loss_value, acc_value, q_val = sess.run(
[train_op, total_loss, accuracy, Q_GLOBAL],
feed_dict={global_step: step})
print(q_val)
elif not FLAGS.dual:
_, loss_value, acc_value = sess.run(
[train_op, total_loss, accuracy],
feed_dict={global_step: step})
elif FLAGS.dual and (step < FLAGS.init_iter
or FLAGS.min_batch_size != -1):
_, _, loss_value, acc_value1, acc_value2, drate = sess.run(
[
train_op1, train_op2, total_loss1, accuracy1,
accuracy2, disagree_rate
],
feed_dict={global_step: step})
else:
#loss_value, acc_value1, acc_value2, drate = (0,0,0,0)
img, lbl, us, loss_value, acc_value1, acc_value2, drate = sess.run(
[
images, labels, update_step, total_loss1, accuracy1,
accuracy2, disagree_rate
],
feed_dict={global_step: step})
rel_img = img[us == 1]
rel_lbl = lbl[us == 1]
if trainable_buffer_img is None:
trainable_buffer_img = rel_img
trainable_buffer_lbl = rel_lbl
else:
print(np.shape(trainable_buffer_lbl), np.shape(rel_lbl))
trainable_buffer_img = np.vstack(
(trainable_buffer_img, rel_img))
trainable_buffer_lbl = np.hstack(
(trainable_buffer_lbl, rel_lbl))
if trainable_buffer_img.shape[0] >= FLAGS.batch_size:
batch_img = trainable_buffer_img[:FLAGS.batch_size]
batch_lbl = trainable_buffer_lbl[:FLAGS.batch_size]
_, _, loss_value, acc_value1, acc_value2, drate = sess.run(
[
train_op1, train_op2, total_loss1, accuracy1,
accuracy2, disagree_rate
],
feed_dict={
global_step: step,
images: batch_img,
labels: batch_lbl
})
trainable_buffer_img = trainable_buffer_img[FLAGS.
batch_size:]
trainable_buffer_lbl = trainable_buffer_lbl[FLAGS.
batch_size:]
#_, loss_value, acc_value2, drate = sess.run([train_op2, total_loss2, accuracy2, disagree_rate], feed_dict={global_step: step})
duration = time.time() - start_time
assert not np.isnan(loss_value), 'Model diverged with loss = NaN'
if step % 1 == 0:
num_examples_per_step = FLAGS.batch_size
examples_per_sec = num_examples_per_step / duration
sec_per_batch = float(duration)
if not FLAGS.dual:
format_str = (
'%s: step %d, loss = %.3f, acc = %.3f (%.1f examples/sec; %.3f '
'sec/batch)')
print(format_str %
(datetime.now(), step, loss_value, acc_value,
examples_per_sec, sec_per_batch))
else:
format_str = (
'%s: step %d, loss = %.3f, acc1 = %.3f, acc2 = %.3f, disagree_rate = %.3f (%.1f examples/sec; %.3f '
'sec/batch)')
print(format_str %
(datetime.now(), step, loss_value, acc_value1,
acc_value2, drate, examples_per_sec, sec_per_batch))
# Loss only actually evaluated every 100 steps?
if step % 100 == 0:
summary_str = sess.run(summary_op)
summary_writer.add_summary(summary_str, step)
if step % 200 == 0 or (step + 1) == num_steps:
saver.save(sess, checkpoint_path, global_step=step)
def main(argv=None):
with tf.Graph().as_default():
serialized_tf_example = tf.placeholder(tf.string, name='tf_example')
feature_configs = {
'image/encoded': tf.FixedLenFeature(shape=[], dtype=tf.string),
}
tf_example = tf.parse_example(serialized_tf_example, feature_configs)
jpegs = tf_example['image/encoded']
images = tf.map_fn(preproc_jpeg, jpegs, dtype=tf.float32)
label_list = AGE_LIST if FLAGS.class_type == 'age' else GENDER_LIST
nlabels = len(label_list)
config = tf.ConfigProto(allow_soft_placement=True)
with tf.Session(config=config) as sess:
model_fn = select_model(FLAGS.model_type)
logits = model_fn(nlabels, images, 1, False)
softmax_output = tf.nn.softmax(logits)
values, indices = tf.nn.top_k(
softmax_output, 2 if FLAGS.class_type == 'age' else 1)
class_tensor = tf.constant(label_list)
table = tf.contrib.lookup.index_to_string_table_from_tensor(
class_tensor)
classes = table.lookup(tf.to_int64(indices))
requested_step = FLAGS.requested_step if FLAGS.requested_step else None
checkpoint_path = '%s' % (FLAGS.model_dir)
model_checkpoint_path, global_step = get_checkpoint(
checkpoint_path, requested_step, FLAGS.checkpoint)
saver = tf.train.Saver()
saver.restore(sess, model_checkpoint_path)
print('Restored model checkpoint %s' % model_checkpoint_path)
output_path = os.path.join(
tf.compat.as_bytes(FLAGS.output_dir),
tf.compat.as_bytes(str(FLAGS.model_version)))
print('Exporting trained model to %s' % output_path)
builder = tf.saved_model.builder.SavedModelBuilder(output_path)
# Build the signature_def_map.
classify_inputs_tensor_info = tf.saved_model.utils.build_tensor_info(
serialized_tf_example)
classes_output_tensor_info = tf.saved_model.utils.build_tensor_info(
classes)
scores_output_tensor_info = tf.saved_model.utils.build_tensor_info(
values)
classification_signature = (
tf.saved_model.signature_def_utils.build_signature_def(
inputs={
tf.saved_model.signature_constants.CLASSIFY_INPUTS:
classify_inputs_tensor_info
},
outputs={
tf.saved_model.signature_constants.CLASSIFY_OUTPUT_CLASSES:
classes_output_tensor_info,
tf.saved_model.signature_constants.CLASSIFY_OUTPUT_SCORES:
scores_output_tensor_info
},
method_name=tf.saved_model.signature_constants.
CLASSIFY_METHOD_NAME))
predict_inputs_tensor_info = tf.saved_model.utils.build_tensor_info(
jpegs)
prediction_signature = (
tf.saved_model.signature_def_utils.build_signature_def(
inputs={'images': predict_inputs_tensor_info},
outputs={
'classes': classes_output_tensor_info,
'scores': scores_output_tensor_info
},
method_name=tf.saved_model.signature_constants.
PREDICT_METHOD_NAME))
legacy_init_op = tf.group(tf.tables_initializer(),
name='legacy_init_op')
builder.add_meta_graph_and_variables(
sess, [tf.saved_model.tag_constants.SERVING],
signature_def_map={
'predict_images':
prediction_signature,
tf.saved_model.signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY:
classification_signature,
},
legacy_init_op=legacy_init_op)
builder.save()
print('Successfully exported model to %s' % FLAGS.output_dir)
def main(argv=None): # pylint: disable=unused-argument
with tf.Session() as sess:
label_list = AGE_LIST if FLAGS.class_type == 'age' else GENDER_LIST
nlabels = len(label_list)
print('Executing on %s' % FLAGS.device_id)
model_fn = select_model(FLAGS.model_type)
images = tf.placeholder(tf.float32, [None, RESIZE_FINAL, RESIZE_FINAL, 3])
logits = model_fn(nlabels, images, 1, False)
init = tf.global_variables_initializer()
requested_step = FLAGS.requested_step if FLAGS.requested_step else None
checkpoint_path = '%s' % (FLAGS.model_dir)
model_checkpoint_path, global_step = get_checkpoint(checkpoint_path, requested_step, FLAGS.checkpoint)
saver = tf.train.Saver()
saver.restore(sess, model_checkpoint_path)
softmax_output = tf.nn.softmax(logits)
coder = ImageCoder()
files = []
if FLAGS.face_detection_model:
print('Using face detector %s' % FLAGS.face_detection_model)
face_detect = FaceDetector(FLAGS.face_detection_model)
face_files, rectangles = face_detect.run(FLAGS.filename)
files += face_files
# Support a batch mode if no face detection model
if len(files) == 0:
files.append(FLAGS.filename)
# If it happens to be a list file, read the list and clobber the files
if one_of(FLAGS.filename, ('csv', 'tsv', 'txt')):
files = batchlist(FLAGS.filename)
writer = None
output = None
if FLAGS.target:
print('Creating output file %s' % FLAGS.target)
output = open(FLAGS.target, 'w')
writer = csv.writer(output)
writer.writerow(('file', 'label', 'score'))
for f in files:
image_file = resolve_file(f)
if image_file is None: continue
try:
best_choice = classify(sess, label_list, softmax_output, coder, images, image_file)
if writer is not None:
writer.writerow((f, best_choice[0], '%.2f' % best_choice[1]))
except Exception as e:
print(e)
print('Failed to run image %s ' % image_file)
if output is not None:
output.close()
def predict(self,
image_file=None,
mode=0,
image_bound=None,
use_tf_to_read=True):
model_dir = self.model_list[mode]
class_type = self.class_list[mode]
files = []
config = tf.ConfigProto(allow_soft_placement=True)
with tf.Session(config=config) as sess:
label_list = self.AGE_LIST if class_type == 'age' else self.GENDER_LIST
nlabels = len(label_list)
model_fn = select_model(self.model_type)
with tf.device(self.device_id):
images = tf.placeholder(
tf.float32,
[None, self.RESIZE_FINAL, self.RESIZE_FINAL, 3])
logits = model_fn(nlabels, images, 1, False)
# init = tf.global_variables_initializer()
requested_step = self.requested_step if self.requested_step else None
checkpoint_path = '%s' % (model_dir)
model_checkpoint_path, global_step = get_checkpoint(
checkpoint_path, requested_step, self.checkpoint)
saver = tf.train.Saver()
saver.restore(sess, model_checkpoint_path)
softmax_output = tf.nn.softmax(logits)
coder = ImageCoder()
# Support a batch mode if no face detection model
# if len(files) == 0:
# if (os.path.isdir(filename)):
# for relpath in os.listdir(filename):
# abspath = os.path.join(filename, relpath)
# print(abspath)
# if os.path.isfile(abspath) and any(
# [abspath.endswith('.' + ty) for ty in ('jpg', 'png', 'JPG', 'PNG', 'jpeg')]):
# print(abspath)
# files.append(abspath)
# else:
# files.append(filename)
# # If it happens to be a list file, read the list and clobber the files
# if any([filename.endswith('.' + ty) for ty in ('csv', 'tsv', 'txt')]):
# files = self.list_images(filename)
writer = None
# output = None
# if self.target:
# output = open(self.target, 'w')
# writer = csv.writer(output)
# writer.writerow(('file', 'label', 'score'))
# image_files = list(filter(lambda x: x is not None, [self.resolve_file(f) for f in files]))
# if self.single_look:
# self.classify_many_single_crop(sess, label_list, softmax_output, coder, images, image_files, writer)
#
# else:
# for image_file in filename:
best_choice, second_choice = self.classify_one_multi_crop(
sess,
label_list,
softmax_output,
coder,
images,
image_file,
writer,
image_bound=image_bound,
use_tf_to_read=use_tf_to_read)
# if output is not None:
# output.close()
return best_choice, second_choice
# Open as Session
#config = tf.ConfigProto(allow_soft_placement=True)
#sess = tf.Session()
# Creating different graphs
g1 = tf.Graph()
g2 = tf.Graph()
# AGE MODEL
with g1.as_default():
session = tf.Session(graph=g1)
with session.as_default():
images = tf.placeholder(tf.float32,
[None, RESIZE_FINAL, RESIZE_FINAL, 3])
ageNet = select_model('default')
logits = ageNet(len(AGE_LIST), images, 1, False)
requested_step = 14999
checkpoint_path = '%s' % ('C:/Users/LEE/Desktop/rude-carnie/age_model')
model_checkpoint_path, global_step = get_checkpoint(
checkpoint_path, requested_step, 'checkpoint')
age_saver = tf.train.Saver()
age_saver.restore(session, model_checkpoint_path)
# GENDER MODEL
with g2.as_default():
session2 = tf.Session(graph=g2)
with session2.as_default():
images2 = tf.placeholder(tf.float32,
[None, RESIZE_FINAL, RESIZE_FINAL, 3])
genderNet = select_model('default')
def main(argv=None):
with tf.Graph().as_default():
model_fn = select_model(FLAGS.model_type)
# Open the metadata file and figure out nlabels, and size of epoch
input_file = os.path.join(FLAGS.train_dir, 'md.json')
print(input_file)
with open(input_file, 'r') as f:
md = json.load(f)
images, labels, _ = distorted_inputs(FLAGS.train_dir, FLAGS.batch_size, FLAGS.image_size, FLAGS.num_preprocess_threads)
logits = model_fn(md['nlabels'], images, 1-FLAGS.pdrop, True)
total_loss = loss(logits, labels)
train_op = optimizer(FLAGS.optim, FLAGS.eta, total_loss)
saver = tf.train.Saver(tf.global_variables())
summary_op = tf.summary.merge_all()
sess = tf.Session(config=tf.ConfigProto(
log_device_placement=FLAGS.log_device_placement))
tf.global_variables_initializer().run(session=sess)
# This is total hackland, it only works to fine-tune iv3
if FLAGS.pre_model:
inception_variables = tf.get_collection(
tf.GraphKeys.VARIABLES, scope="InceptionV3")
restorer = tf.train.Saver(inception_variables)
restorer.restore(sess, FLAGS.pre_model)
if FLAGS.pre_checkpoint_path:
if tf.gfile.Exists(FLAGS.pre_checkpoint_path) is True:
print('Trying to restore checkpoint from %s' % FLAGS.pre_checkpoint_path)
restorer = tf.train.Saver()
tf.train.latest_checkpoint(FLAGS.pre_checkpoint_path)
print('%s: Pre-trained model restored from %s' %
(datetime.now(), FLAGS.pre_checkpoint_path))
run_dir = '%s/run-%d' % (FLAGS.train_dir, os.getpid())
checkpoint_path = '%s/%s' % (run_dir, FLAGS.checkpoint)
if tf.gfile.Exists(run_dir) is False:
print('Creating %s' % run_dir)
tf.gfile.MakeDirs(run_dir)
tf.train.write_graph(sess.graph_def, run_dir, 'model.pb', as_text=True)
tf.train.start_queue_runners(sess=sess)
summary_writer = tf.summary.FileWriter(run_dir, sess.graph)
steps_per_train_epoch = int(md['train_counts'] / FLAGS.batch_size)
num_steps = FLAGS.max_steps if FLAGS.epochs < 1 else FLAGS.epochs * steps_per_train_epoch
print('Requested number of steps [%d]' % num_steps)
for step in xrange(num_steps):
start_time = time.time()
_, loss_value = sess.run([train_op, total_loss])
duration = time.time() - start_time
assert not np.isnan(loss_value), 'Model diverged with loss = NaN'
if step % 10 == 0:
num_examples_per_step = FLAGS.batch_size
examples_per_sec = num_examples_per_step / duration
sec_per_batch = float(duration)
format_str = ('%s: step %d, loss = %.3f (%.1f examples/sec; %.3f ' 'sec/batch)')
print(format_str % (datetime.now(), step, loss_value,
examples_per_sec, sec_per_batch))
# Loss only actually evaluated every 100 steps?
if step % 100 == 0:
summary_str = sess.run(summary_op)
summary_writer.add_summary(summary_str, step)
if step % 1000 == 0 or (step + 1) == num_steps:
saver.save(sess, checkpoint_path, global_step=step)
nlabels = len(label_list)
if nlabels > 2:
output[best] = 0
second_best = np.argmax(output)
print('Guess @ 2 %s, prob = %.2f' % (label_list[second_best], output[second_best]))
return best_choice
#调用guessGender前要先初始化
# tf.reset_default_graph()
label_list = GENDER_LIST
nlabels = len(label_list)
print('Executing on %s' % FLAGS.device_id)
model_fn = select_model(FLAGS.model_type)
images = tf.placeholder(tf.float32, [None, RESIZE_FINAL, RESIZE_FINAL, 3])
logits = model_fn(nlabels, images, 1, False)
def guessGender(file_name): # pylint: disable=unused-argument
#检测单张照片的性别
with tf.Session() as sess:
with tf.device(FLAGS.device_id):
init = tf.global_variables_initializer()
requested_step = FLAGS.requested_step if FLAGS.requested_step else None
checkpoint_path = '%s' % (GENDER_MODEL_PATH)
def main(argv=None): # pylint: disable=unused-argument
files = []
if FLAGS.face_detection_model:
print('Using face detector (%s) %s' % (FLAGS.face_detection_type, FLAGS.face_detection_model))
face_detect = face_detection_model(FLAGS.face_detection_type, FLAGS.face_detection_model)
face_files, rectangles = face_detect.run(FLAGS.filename)
print(face_files)
files += face_files
with tf.Session() as sess:
#tf.reset_default_graph()
label_list = AGE_LIST if FLAGS.class_type == 'age' else GENDER_LIST
nlabels = len(label_list)
print('Executing on %s' % FLAGS.device_id)
model_fn = select_model(FLAGS.model_type)
with tf.device(FLAGS.device_id):
images = tf.placeholder(tf.float32, [None, RESIZE_FINAL, RESIZE_FINAL, 3])
logits = model_fn(nlabels, images, 1, False)
init = tf.global_variables_initializer()
requested_step = FLAGS.requested_step if FLAGS.requested_step else None
checkpoint_path = '%s' % (FLAGS.model_dir)
model_checkpoint_path, global_step = get_checkpoint(checkpoint_path, requested_step, FLAGS.checkpoint)
saver = tf.train.Saver()
saver.restore(sess, model_checkpoint_path)
softmax_output = tf.nn.softmax(logits)
coder = ImageCoder()
# Support a batch mode if no face detection model
if len(files) == 0:
files.append(FLAGS.filename)
# If it happens to be a list file, read the list and clobber the files
if one_of(FLAGS.filename, ('csv', 'tsv', 'txt')):
files = batchlist(FLAGS.filename)
writer = None
output = None
if FLAGS.target:
print('Creating output file %s' % FLAGS.target)
output = open(FLAGS.target, 'w')
writer = csv.writer(output)
writer.writerow(('file', 'label', 'score'))
for f in files:
image_file = resolve_file(f)
if image_file is None: continue
try:
best_choice = classify(sess, label_list, softmax_output, coder, images, image_file)
if writer is not None:
writer.writerow((f, best_choice[0], '%.2f' % best_choice[1]))
except Exception as e:
print(e)
print('Failed to run image %s ' % image_file)
if output is not None:
output.close()