def verify(align_imgs):
"""Verify images after align
@input: image after align
@output: distance
"""
g = get_graph()
with g.as_default():
# Get input and output tensors
images_placeholder = g.get_tensor_by_name("input:0")
embeddings = g.get_tensor_by_name("embeddings:0")
phase_train_placeholder = g.get_tensor_by_name("phase_train:0")
# Run forward pass to calculate embeddings
feed_dict = {
images_placeholder: align_imgs,
phase_train_placeholder: False
}
sess = get_session()
emb = sess.run(embeddings, feed_dict=feed_dict)
dist = np.sqrt(np.sum(np.square(np.substract(emb[0, :], emb[1, :]))))
print('distance: %1.4f' % dist)
return '%1.4f' % dist
def init_session():
global _sess
# single session for facenet
_sess = tf.Session(graph=get_graph())
# load_model(sess, model_dir)
exp_model_path = os.path.expanduser(_model_path)
load_model(_sess, exp_model_path)
def init():
# measure time_used for init
start = time.time()
graph = tfgraph.get_graph()
# sess = tf.Session(graph=graph)
sess = tfsession.get_session()
return jsonify(time_used=time.time() - start)
def load_model(sess, model_dir):
"""Load facenet model
"""
# measure time used for loading model
start = time.time()
model_dir_exp = os.path.expanduser(model_dir)
meta_file, ckpt_file = get_model_filenames(model_dir)
meta_file_full_path = os.path.join(model_dir_exp, meta_file)
ckpt_file_full_path = os.path.join(model_dir_exp, ckpt_file)
print('Model meta file: ', meta_file_full_path)
print('Model ckpt file: ', ckpt_file_full_path)
print('Loading models. Waiting...')
from ngface.tfgraph import get_graph
g = get_graph()
with g.as_default():
saver = tf.train.import_meta_graph(meta_file_full_path)
saver.restore(sess, ckpt_file_full_path)
print('Models loaded. time_used: ', time.time() - start)
def verify():
# verify time_used
start = time.time()
# read image from request
img_list = utils.get_images_from_request(request.files, ['img1', 'img2'])
np_images = load_and_align_data(img_list)
dist_str = ''
dist = 0
graph = tfgraph.get_graph()
with graph.as_default():
# Get input and output tensors
images_placeholder = graph.get_tensor_by_name("input:0")
embeddings = graph.get_tensor_by_name("embeddings:0")
phase_train_placeholder = graph.get_tensor_by_name("phase_train:0")
# Run forward pass to calculate embeddings
feed_dict = {
images_placeholder: np_images,
phase_train_placeholder: False
}
sess = tfsession.get_session()
emb = sess.run(embeddings, feed_dict=feed_dict)
# nrof_images = len(args.image_files)
dist = np.sqrt(np.sum(np.square(np.subtract(emb[0, :], emb[1, :]))))
dist_str = '%1.4f' % dist
print('Distance: ', dist_str)
threshold = 0.9
print('Threshold: ', threshold)
time_used = time.time() - start
return jsonify(is_same_person=str(dist < threshold),
time_used=str(time_used))