def main(matcher_path, test_path):
m_trackers_paths = glob.glob(matcher_path + '/*')
t_trackers_paths = glob.glob(test_path + '/*')
tracker_manager = TrackerManager('test')
matcher = FaissMatcher()
preprocessor = Preprocessor()
align_preprocessor = Preprocessor(algs=align_and_crop)
face_rec_graph_face = FaceGraph()
face_extractor = FacenetExtractor(face_rec_graph_face,
model_path=Config.FACENET_DIR)
detector = MTCNNDetector(face_rec_graph_face)
# create matcher
print('Creating matcher ...')
for m_dir in m_trackers_paths:
print('Processing ' + m_dir)
face_id = m_dir.split('/')[-1]
embs, labels = extract_embs(m_dir, preprocessor, face_extractor, None)
face_id_labels = [face_id for i in range(len(labels))]
matcher.update(embs, face_id_labels)
# create tracker
print('Creating trackers')
for t_dir in t_trackers_paths:
print('Processing ' + t_dir)
embs, _ = extract_embs(t_dir, preprocessor, face_extractor, None)
track_id = int(t_dir.split('/')[-1])
first_emb = embs.pop()
face_info = FaceInfo(None, first_emb, None, None, None, None)
tracker_manager.current_trackers[track_id] = Tracker(
track_id, face_info, None)
for emb in embs:
face_info = FaceInfo(None, emb, None, None, None, None)
tracker_manager.current_trackers[track_id].update(face_info, None)
len(tracker_manager.current_trackers)
# test matching
print('Test matching ...')
for fid in tracker_manager.current_trackers:
print('Processing: ' + str(fid))
tops = tracker_manager.recognize_current_tracker(fid, matcher, None)
print('Track_id {}, recognize: {}'.format(fid, tops))
def give_this_id_10_closest_ids():
# init matcher
with open('../data/top10querylog.txt', 'a') as f:
f.write('TOP10 QUERY IS BEING IN PROCESS !!!\n\n')
global querying_top10_image_ids_queue
global mongodb_faceinfo
global mongodb_dashinfo
embs = []
labels = []
cursors = mongodb_dashinfo.find({})
unique_labels = [cursor['represent_image_id'] for cursor in cursors]
cursors = mongodb_faceinfo.find({'image_id': {'$in': unique_labels}})
for cursor in cursors:
embs.append(np.array(cursor['embedding']))
labels.append(cursor['image_id'])
nof_registered_image_ids = len(labels)
matcher = FaissMatcher()
matcher.fit(embs, labels)
with open('../data/top10querylog.txt', 'a') as f:
f.write('MATCHER BUILT!!!\n\n')
while True:
if nof_registered_image_ids < mongodb_dashinfo.find({}).count():
nof_registered_image_ids = mongodb_dashinfo.find({}).count()
print('[Query TOP10] Update new registered image_id ...')
cursors = mongodb_dashinfo.find({
'represent_image_id': {
'$nin': labels
}
})
unique_labels = [cursor['represent_image_id'] for cursor in cursors]
cursors = mongodb_faceinfo.find({
'image_id': {
'$in': unique_labels
}
})
adding_embs = []
adding_labels = []
for cursor in cursors:
adding_embs.append(np.array(cursor['embedding']))
adding_labels.append(cursor['image_id'])
old_embs = embs
old_labels = labels
embs = old_embs + adding_embs
labels = old_labels + adding_labels
matcher.update(adding_embs, adding_labels)
if not len(querying_top10_image_ids_queue) == 0:
lock.acquire()
queue_data = querying_top10_image_ids_queue.pop()
lock.release()
results = {}
session_id = queue_data['sessionId']
image_id = queue_data['imageId']
print('[Query TOP10] image_id: ' + image_id)
with open('../data/top10querylog.txt', 'a') as f:
f.write('image_id: ' + image_id + '\n')
cursors = mongodb_faceinfo.find({'image_id': image_id})
if cursors.count() == 0:
print('[Query TOP10] THIS QUERY IMAGE ID HAS YET TO REGISTER')
with open('../data/top10querylog.txt', 'a') as f:
f.write('THIS QUERY IMAGE ID HAS YET TO REGISTER\n')
face_id = mongodb_dashinfo.find({
'represent_image_id': image_id
})[0]['face_id']
unique_labels = [
cursor['represent_image_id']
for cursor in mongodb_dashinfo.find({
'face_id': face_id
})
]
for label in unique_labels:
results[label] = '0'
else:
query_emb = cursors[0]['embedding']
dists, inds = matcher._classifier.search(
np.array(query_emb).astype('float32'), k=15)
dists = np.squeeze(dists)
inds = np.squeeze(inds)
top_match_ids = [labels[idx] for idx in inds]
for i, top_match_id in enumerate(top_match_ids):
if i < 11 and top_match_id != image_id:
results[top_match_id] = str(dists[i])
msg_results = {
'actionType': 'getNearest',
'sessionId': session_id,
'data': {
'images': results
}
}
with open('../data/top10querylog.txt', 'a') as f:
f.write('Result: \n{}\n\n'.format(results))
print('[Query TOP10] Result: \n{}'.format(results))
rabbit_mq.send_with_exchange(Config.Queues.ACTION_RESULT,
session_id, json.dumps(msg_results))
# Those cmt for querying tracker from the image ids tracker
# query_track_id = int(image_id.split('_')[0])
# query_embs = [cursor['embedding'] for cursor in mongodb_faceinfo.find({'track_id': query_track_id})]
# for emb in query_embs:
# predict_id, _, min_dist = matcher.match(emb, return_min_dist=True)
# if not predict_id in predicted_dict:
# predicted_dict[predict_id] = []
# predicted_dict[predict_id].append(min_dist)
# avg_predicted_dict = {pid: sum(predicted_dict[pid])/float(len(predicted_dict[pid]))
# for pid in predicted_dict}
# sorted_predicted_ids = sorted(avg_predicted_dict.items(), key=lambda kv: kv[1])
# with open('../data/query_top10_log.txt') as f:
# f.write('Query IMAGE_ID: ' + image_id + '\n')
# f.write('Results: {} \n\n'.format(sorted_predicted_ids))
# str_results = []
# for closest_id, dist in sorted_predicted_ids:
# str_results.append(Config.Rabbit.INTRA_SEP.join([closest_id, str(dist)]))
# result_msg = Config.Rabbit.INTER_SEP.join(str_results)
else:
time.sleep(1)
class FindSimilarFaceThread(Thread):
'''
Find similar faces from dashboard
'''
def __init__(self, **args):
self.nrof_closest_faces = args.get('nrof_closest_faces', 10)
self.database = args.get('database')
self.rabbit_mq = args.get('rabbit_mq')
self.event = threading.Event()
self.setup_matcher()
super(FindSimilarFaceThread, self).__init__()
def join(self, timeout=None):
print('Find similar joint')
self.event.set()
super(FindSimilarFaceThread, self).join()
def setup_matcher(self):
with open('../data/top10querylog.txt', 'a') as f:
f.write('TOP10 QUERY IS BEING IN PROCESS !!!\n\n')
'''
self.embs = []
self.labels = []
cursors = self.mongo.mongodb_dashinfo.find({})
unique_labels = [cursor['represent_image_id'] for cursor in cursors]
cursors = self.mongo.mongodb_faceinfo.find({'image_id': {'$in': unique_labels}})
for cursor in cursors:
self.embs.append(np.array(cursor['embedding']))
self.labels.append(cursor['image_id'])
self.nof_registered_image_ids = len(self.labels)
'''
self.labels, self.embs = self.database.get_labels_and_embs_dashboard()
self.matcher = FaissMatcher()
self.matcher.fit(self.embs, self.labels)
with open('../data/top10querylog.txt', 'a') as f:
f.write('MATCHER BUILT!!!\n\n')
def run(self):
while not self.event.is_set():
# first update check for new faces in dashboard
'''
if self.nof_registered_image_ids < self.mongo.mongodb_dashinfo.find({}).count():
self.nof_registered_image_ids = self.mongo.mongodb_dashinfo.find({}).count()
print('[Query TOP10] Update new registered image_id ...')
cursors = self.mongo.mongodb_dashinfo.find({'represent_image_id': {'$nin': self.labels}})
unique_labels = [cursor['represent_image_id'] for cursor in cursors]
cursors = self.mongo.mongodb_faceinfo.find({'image_id': {'$in': unique_labels}})
adding_embs = []
adding_labels = []
for cursor in cursors:
adding_embs.append(np.array(cursor['embedding']))
adding_labels.append(cursor['image_id'])
'''
adding_labels, adding_embs = self.database.get_labels_and_embs_dashboard(
self.labels)
if len(adding_labels) > 0:
old_embs = self.embs
old_labels = self.labels
self.embs = old_embs + adding_embs
self.labels = old_labels + adding_labels
print('Find similar', len(adding_labels))
self.matcher.update(adding_embs, adding_labels)
# get new query from from queue, why not just trigger
action_msg = self.rabbit_mq.receive_str(Config.Queues.ACTION)
if action_msg is not None:
return_dict = json.loads(action_msg)
print('Receive: {}'.format(return_dict))
if return_dict['actionType'] == 'getNearest':
data = return_dict['data']
results = {}
session_id = data['sessionId']
image_id = data['imageId']
print('[Query TOP10] image_id: ' + image_id)
with open('../data/top10querylog.txt', 'a') as f:
f.write('image_id: ' + image_id + '\n')
cursors = self.database.mongodb_faceinfo.find(
{'image_id': image_id})
if cursors.count() == 0:
print(
'[Query TOP10] THIS QUERY IMAGE ID HAS YET TO REGISTER'
)
with open('../data/top10querylog.txt', 'a') as f:
f.write(
'THIS QUERY IMAGE ID HAS YET TO REGISTER\n')
face_id = self.database.mongodb_dashinfo.find(
{'represent_image_id': image_id})[0]['face_id']
unique_labels = [
cursor['represent_image_id']
for cursor in self.database.mongodb_dashinfo.find(
{'face_id': face_id})
]
for label in unique_labels:
results[label] = '0'
else:
query_emb = cursors[0]['embedding']
embs = np.array(query_emb).astype('float32').reshape(
(-1, 128))
dists, inds = self.matcher._classifier.search(embs,
k=15)
dists = np.squeeze(dists)
inds = np.squeeze(inds)
top_match_ids = [self.labels[idx] for idx in inds]
for i, top_match_id in enumerate(top_match_ids):
if i < 11 and top_match_id != image_id:
results[top_match_id] = str(dists[i])
msg_results = {
'actionType': 'getNearest',
'sessionId': session_id,
'data': {
'images': results
}
}
with open('../data/top10querylog.txt', 'a') as f:
f.write('Result: \n{}\n\n'.format(results))
print('[Query TOP10] Result: \n{}'.format(results))
self.rabbit_mq.send_with_exchange(
Config.Queues.ACTION_RESULT, session_id,
json.dumps(msg_results))
else:
time.sleep(1)