async def detection_loop(preload):
# =================== FD MODEL ====================
detector = face_detector.DetectorModel(preload)
embedding_threshold = preload.embedding_threshold
rate = preload.max_frame_rate
loop = asyncio.get_running_loop()
def retina_deal(pair):
address, frame = pair
res = []
for img, box in detector.get_all_boxes(frame, save_img=False):
res.append(box.tolist())
if box[4] > embedding_threshold:
try:
suspicion_face_queue.put_nowait((address, img))
except:
pass
return (address, frame) if res == [] else (address, res, frame)
while True:
start_time = loop.time()
frame_list = [retina_deal(pair) for pair in frame_queue.get()]
print(colored(f'Detection cost: {loop.time() - start_time}', 'red'),
flush=True)
upstream_queue.put(frame_list)
for _ in range(int((loop.time() - start_time) * rate)):
upstream_queue.put(frame_queue.get())
async def detection_loop(preload):
# =================== FD MODEL ====================
detector = face_detector.DetectorModel(preload)
rate = preload.max_frame_rate
loop = asyncio.get_running_loop()
while True:
start_time = loop.time()
head_frame_list = frame_queue.get()
for (ip_address, head_frame) in head_frame_list:
for img, box in detector.get_all_boxes(head_frame, save_img=False):
try:
suspicion_face_queue.put_nowait((ip_address, img))
except Exception as e:
pass
box = box.astype(int)
cv2.rectangle(head_frame, (box[0], box[1]), (box[2], box[3]),
[255, 255, 0], 2)
upstream_queue.put((ip_address, head_frame))
print(colored(loop.time() - start_time, 'red'), flush=True)
for i in range(int((loop.time() - start_time) * rate + 1)):
for item in frame_queue.get():
upstream_queue.put(item)
async def detection_loop(preload, address):
# =================== FD MODEL ====================
detector = face_detector.DetectorModel(preload)
rate_time = 1 / preload.max_frame_rate
embedding_threshold = preload.embedding_threshold
loop = asyncio.get_running_loop()
camera = ipc.XMIPCamera(address.encode('UTF-8'), 34567, b"admin", b"")
camera.start()
while True:
start_time = loop.time()
frame = camera.frame(rows=540, cols=960)
for img, box in detector.get_all_boxes(frame, save_img=False):
if box[4] > embedding_threshold:
try:
suspicion_face_queue.put_nowait((address, img))
except Exception as _:
pass
# print(box[4])
box = box.astype(np.int)
cv2.rectangle(frame, (box[0], box[1]), (box[2], box[3]),
[255, 255, 0], 2)
upstream_queue.put((address, frame))
# print(colored(loop.time() - start_time, 'red'), flush=True)
restime = rate_time - loop.time() + start_time
if restime > 0:
await asyncio.sleep(restime)
async def detection_loop(preload, frame_queue):
# =================== FD MODEL ====================
detector = face_detector.DetectorModel(preload)
ip_address = preload.ip_address
loop = asyncio.get_running_loop()
while True:
start_time = loop.time()
head_frame = frame_queue.get()
# tracker = cv2.MultiTracker_create()
# t_box = []
for img, box in detector.get_all_boxes(head_frame, save_img=False):
try:
if box[4] > 0.98:
suspicion_face_queue.put_nowait(img)
except Exception as e:
pass
box = box.astype(int)
cv2.rectangle(
head_frame, (box[0], box[1]), (box[2], box[3]), [255, 255, 0], 2)
# t_box.append(box[:4]/2)
print(colored(loop.time()-start_time, 'blue'))
# head_frame = cv2.resize(head_frame, (960, 540), cv2.INTER_AREA)
# for item in t_box:
# tracker.add(cv2.TrackerMedianFlow_create(), head_frame, tuple(item))
upstream_frame_queue.put((ip_address, head_frame))
for i in range(int((loop.time() - start_time) * 25 + 1)):
body_frame = frame_queue.get()
# ok, tricker_boxes = tracker.update(body_frame)
# if ok:
# for box in tricker_boxes:
# box = box.astype(int)
# cv2.rectangle(body_frame, (box[0], box[1]),
# (box[2], box[3]), [255, 255, 0], 2)
upstream_frame_queue.put((ip_address, body_frame))
# await sio.emit('frame_data', encode_image(body_frame), namespace='/remilia')
# end_time = loop.time()
# print(colored(loop.time()-track_time, 'red'))
sys.stdout.flush()
async def detection_loop(preload):
reciprocal_of_max_frame_rate = 1/preload.max_frame_rate
address = preload.address
xmcp = ipc.XMIPCamera(address.encode('UTF-8'), 34567, b"admin", b"")
xmcp.start()
# =================== FD MODEL ====================
detector = face_detector.DetectorModel(preload)
rate = preload.max_frame_rate
loop = asyncio.get_running_loop()
predictor = dlib.shape_predictor(preload.face_landmark_path)
(lStart, lEnd) = face_utils.FACIAL_LANDMARKS_IDXS["left_eye"]
(rStart, rEnd) = face_utils.FACIAL_LANDMARKS_IDXS["right_eye"]
# =================== ETERNAL LOOP ====================
while True:
start_time = loop.time()
try:
head_frame = xmcp.frame(rows=540, cols=960)
for img, box, points in detector.get_boxes_and_landmarks(head_frame):
box = box.astype(int)
# nimg = cv2.resize(
# head_frame[box[1]:box[3], box[0]:box[2]], (112, 112))
shape = predictor(img, dlib.rectangle(0, 0, 112, 112))
# shape = predictor(head_frame, dlib.rectangle(
# box[0], box[1], box[2], box[3]))
# shape = face_utils.shape_to_np(shape) + 50
shape = face_utils.shape_to_np(shape)
reprojectdst, euler_angle = get_head_pose(shape)
# for (x, y) in shape:
# cv2.circle(head_frame, (x, y), 2, (0, 0, 255), -1)
# for start, end in line_pairs:
# cv2.line(
# head_frame, reprojectdst[start], reprojectdst[end], (0, 0, 255))
# =================== EYES ====================
shape = shape * 10
leftEye = shape[lStart:lEnd]
rightEye = shape[rStart:rEnd]
leftEAR = eye_aspect_ratio(leftEye)
rightEAR = eye_aspect_ratio(rightEye)
# head_frame = np.zeros([540, 960, 3])
result_queue.put((euler_angle, shape, leftEAR, rightEAR))
except Exception as e:
print(e)
pass
print(colored(loop.time()-start_time, 'red'), flush=True)
restime = reciprocal_of_max_frame_rate - loop.time() + start_time
if restime > 0:
await asyncio.sleep(restime)
from sklearn.preprocessing import label_binarize
import imageGenerator
from sklearn import metrics
from helper import read_pkl_model, start_up_init, get_dataset, get_image_paths_and_labels
import face_embedding
import face_detector
# =================== ARGS ====================
os.environ["MXNET_CUDNN_AUTOTUNE_DEFAULT"] = "0"
args = start_up_init()
args.retina_model = './model/M25'
args.scales = [0.5]
# =================== MODEL CLASS ====================
detector = face_detector.DetectorModel(args)
arcface = face_embedding.EmbeddingModel(args)
# =================== LOAD DATASET ====================.
dir_train = './Temp/train.npy'
data_train = './Temp/train_data'
dataset_train = get_dataset(data_train)
paths_train, labels_train = get_image_paths_and_labels(dataset_train)
try:
train_emb_array = np.load(dir_train)
train_emb_array = imageGenerator.generator()
except OSError:
if not os.path.exists('./Temp/raw/'):
os.makedirs('./Temp/raw/')
detector.get_all_boxes_from_path(paths_train, save_img=True)
dataset_train = get_dataset(data_train)