def Predict_Dense():
# 从摄像头中读取视频流
video_capture = cv2.VideoCapture(0)
check_weight(DENSE_WEIGHT_PATH)
model = dense_models()
# 加载权重
model.load_weights(DENSE_WEIGHT_PATH)
while True: # 读取每一帧
ret, frame = video_capture.read() # 0.01s
small_frame = cv2.resize(frame, (0, 0), fx=0.75, fy=0.75)
rgb_frame = small_frame[:, :, ::-1]
face_locations = face_recognition.face_locations(rgb_frame) # 0.17s
if face_locations == []:
continue
face_encodings = face_recognition.face_encodings(
rgb_frame, face_locations) # 找到了当前帧中所有面部的编码 #0.51s
for face_location in face_locations: # 画矩形时间：不足0.001s
top, right, bottom, left = face_location
cv2.rectangle(small_frame, (left, top), (right, bottom),
(0, 0, 255), 2)
out = model.predict(np.array(face_encodings)) # 神经网络预测时间：0.001s
for o in out:
print("Welcome people：{}".format(o.argmax() + 1))
cv2.imshow('Video', small_frame) # 展示一帧时间：不足0.001s
if cv2.waitKey(1) & 0xFF == ord('q'):
break
video_capture.release()
cv2.destroyAllWindows
def test():
check_weight()
model = dense_models()
model.load_weights(WEIGHT_PATH)
data = []
test_x, test_y = get_encodings(False)
data.append(test_x)
data = np.array(data).astype(np.float32).reshape(-1,128)
out = model.test_on_batch(data,test_y)
print("loss = {}, accuracy = {}".format(out[0],out[1]))
def train():
batch_x, batch_y = get_encodings()
model =dense_models()
if os.path.exists(WEIGHT_PATH):
model.load_weights(WEIGHT_PATH)
tb = TensorBoard(log_dir=LOG_PATH, write_images=1)
call_list = [tb]
model.fit(x=batch_x, y=batch_y, batch_size=batch_size, epochs = epochs, verbose=2,callbacks=call_list) # callback回调函数会在训练适当时机被调用
model.save_weights(WEIGHT_PATH, True)