def main(args):
frozen_model = args.frozen_model
svc_model = args.svc_model
mtcnn_model = args.mtcnn_model
with open(svc_model, 'rb') as infile:
(model, class_names) = pickle.load(infile)
class_names_print = ['Lai Ngoc Thang Long', 'Luong Trong Tri', \
'Ngo Song Viet Hoang', 'Nguyen Ba Đuc', 'Nguyen Hoang Son',\
'Nguyen Dinh Dung', 'Nguyen Dinh Hung', 'Tran Quang Ha', 'Vu Minh Thanh']
detector = MTCNN(mtcnn_model)
cap = cv2.VideoCapture(0)
while (True):
# Capture frame-by-frame
ret, frame = cap.read()
# Our operations on the frame come here
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2RGB)
faces, bboxs = cut_face(gray, detector)
if faces:
for i in range(len(faces)):
face = faces[i]
bounding_box = bboxs[i]
index, score = facenet_svc(frozen_model, model, class_names,
face)
if score >= 0.5:
cv2.rectangle(frame, (bounding_box[0], bounding_box[1]),
(bounding_box[2], bounding_box[3]),
(0, 155, 255), 2)
cv2.putText(frame,
class_names_print[index] + " : " + str(score),
(bounding_box[0], bounding_box[1]),
cv2.FONT_ITALIC, 0.8, (0, 255, 0), 1,
cv2.LINE_AA)
cv2.imshow('frame', frame)
if cv2.waitKey(1) & 0xFF == ord('q'):
break
# When everything done, release the capture
cap.release()
cv2.destroyAllWindows()
def main():
meta_file = './models/model8/model.meta'
ckpt_file = './models/model8/model.ckpt-63234'
image_size = 112
with tf.Graph().as_default():
with tf.Session() as sess:
print('Loading feature extraction model.')
saver = tf.train.import_meta_graph(meta_file)
saver.restore(tf.get_default_session(), ckpt_file)
graph = tf.get_default_graph()
images_placeholder = graph.get_tensor_by_name('image_batch:0')
phase_train_placeholder = graph.get_tensor_by_name('phase_train:0')
landmark_L1 = graph.get_tensor_by_name('landmark_L1:0')
landmark_L2 = graph.get_tensor_by_name('landmark_L2:0')
landmark_L3 = graph.get_tensor_by_name('landmark_L3:0')
landmark_L4 = graph.get_tensor_by_name('landmark_L4:0')
landmark_L5 = graph.get_tensor_by_name('landmark_L5:0')
landmark_total = [
landmark_L1, landmark_L2, landmark_L3, landmark_L4, landmark_L5
]
cap = cv2.VideoCapture(0)
mtcnn = MTCNN()
while True:
ret, image = cap.read()
height, width, _ = image.shape
if not ret: break
boxes = mtcnn.predict(image)
for box in boxes:
score = box[4]
x1, y1, x2, y2 = (box[:4] + 0.5).astype(np.int32)
w = x2 - x1 + 1
h = y2 - y1 + 1
size = int(max([w, h]) * 1.1)
cx = x1 + w // 2
cy = y1 + h // 2
x1 = cx - size // 2
x2 = x1 + size
y1 = cy - size // 2
y2 = y1 + size
dx = max(0, -x1)
dy = max(0, -y1)
x1 = max(0, x1)
y1 = max(0, y1)
edx = max(0, x2 - width)
edy = max(0, y2 - height)
x2 = min(width, x2)
y2 = min(height, y2)
cropped = image[y1:y2, x1:x2]
if (dx > 0 or dy > 0 or edx > 0 or edy > 0):
cropped = cv2.copyMakeBorder(cropped, dy, edy, dx, edx,
cv2.BORDER_CONSTANT, 0)
cropped = cv2.resize(cropped, (image_size, image_size))
input = cv2.resize(cropped, (image_size, image_size))
input = cv2.cvtColor(input, cv2.COLOR_BGR2RGB)
input = input.astype(np.float32) / 256.0
input = np.expand_dims(input, 0)
feed_dict = {
images_placeholder: input,
phase_train_placeholder: False
}
cv2.rectangle(image, (x1, y1), (x2, y2), (0, 255, 0))
pre_landmarks = sess.run(landmark_total,
feed_dict=feed_dict)
pre_landmark = pre_landmarks[0]
pre_landmark = pre_landmark.reshape(-1, 2) * [size, size]
for (x, y) in pre_landmark.astype(np.int32):
cv2.circle(image, (x1 + x, y1 + y), 1, (0, 0, 255))
cv2.imshow('0', image)
if cv2.waitKey(10) == 27:
break
def getLandMarks(img_path):
image_size = 112
lite_filename = 'deploy/landmarks68/pfld_landmarks.tflite'
# load TFLite model and allocate tensors
interpreter = tf.lite.Interpreter(model_path=lite_filename)
interpreter.allocate_tensors()
# get input and output tensors.
input_details = interpreter.get_input_details()
output_details = interpreter.get_output_details()
with tf.Graph().as_default():
with tf.Session() as sess:
mtcnn = MTCNN()
image = cv2.imread(img_path)
height, width, _ = image.shape
boxes, landmarks_mtcnn = mtcnn.predict(image)
if len(landmarks_mtcnn) > 0:
landmarks_mtcnn = np.asarray(landmarks_mtcnn).reshape(len(boxes), 10)
for box in boxes:
score = box[4]
x1, y1, x2, y2 = (box[:4] + 0.5).astype(np.int32)
# mtcnn boundding box fine adjustment
dta_y = (min(landmarks_mtcnn[0][5], landmarks_mtcnn[0][6]) - y1) / 2.6
y1 = int(y1 + dta_y)
y2 = int(y2 + dta_y / 2)
w = x2 - x1 + 1
h = y2 - y1 + 1
size = int(max([w, h]) * 1.05)
cx = x1 + w // 2
cy = y1 + h // 2
x1 = cx - size // 2
x2 = x1 + size
y1 = cy - size // 2
y2 = y1 + size
dx = max(0, -x1)
dy = max(0, -y1)
x1 = max(0, x1)
y1 = max(0, y1)
edx = max(0, x2 - width)
edy = max(0, y2 - height)
x2 = min(width, x2)
y2 = min(height, y2)
cropped = image[y1:y2, x1:x2]
if (dx > 0 or dy > 0 or edx > 0 or edy > 0):
cropped = cv2.copyMakeBorder(cropped, dy, edy, dx, edx,
cv2.BORDER_CONSTANT, 0)
cropped = cv2.resize(cropped, (image_size, image_size))
input = cv2.resize(cropped, (image_size, image_size))
input = cv2.cvtColor(input, cv2.COLOR_BGR2RGB)
input = input.astype(np.float32) / 256.0
input = np.expand_dims(input, 0)
interpreter.set_tensor(input_details[0]['index'], input)
interpreter.invoke()
pre_landmarks = interpreter.get_tensor(output_details[0]['index'])
pre_landmark = pre_landmarks[0]
pre_landmark = pre_landmark.reshape(-1, 2) * [size, size]
return pre_landmark, x1, y1
def main():
#meta_file = './models2/model4/model.meta'
#ckpt_file = './models2/model4/model.ckpt-61'
meta_file = './models1/model_test/model.meta'
ckpt_file = './models1/model_test/model.ckpt-13'
image_size = 112
with tf.Graph().as_default():
with tf.compat.v1.Session() as sess:
print('Loading feature extraction model.')
saver = tf.compat.v1.train.import_meta_graph(meta_file)
saver.restore(tf.compat.v1.get_default_session(), ckpt_file)
graph = tf.compat.v1.get_default_graph()
images_placeholder = graph.get_tensor_by_name('image_batch:0')
phase_train_placeholder = graph.get_tensor_by_name('phase_train:0')
landmarks = graph.get_tensor_by_name('pfld_inference/fc/BiasAdd:0')
# landmark_L1 = graph.get_tensor_by_name('landmark_L1:0')
# landmark_L2 = graph.get_tensor_by_name('landmark_L2:0')
# landmark_L3 = graph.get_tensor_by_name('landmark_L3:0')
# landmark_L4 = graph.get_tensor_by_name('landmark_L4:0')
# landmark_L5 = graph.get_tensor_by_name('landmark_L5:0')
# landmark_total = [landmark_L1, landmark_L2, landmark_L3, landmark_L4, landmark_L5]
cap = cv2.VideoCapture(0)
cap_width = 640
cap_height = 480
cap.set(3, cap_width)
cap.set(4, cap_height)
mtcnn = MTCNN()
while True:
ret, image = cap.read()
height, width, _ = image.shape
if not ret:
break
boxes = mtcnn.predict(image)
for box in boxes:
# score = box[4]
x1, y1, x2, y2 = (box[:4] + 0.5).astype(np.int32)
w = x2 - x1 + 1
h = y2 - y1 + 1
size = int(max([w, h]) * 1.1)
# center point
cx = x1 + w // 2
cy = y1 + h // 2
# square
x1 = cx - size // 2 #可能为负值
y1 = cy - size // 2 #可能为负值
x2 = x1 + size # 可能大于宽度
y2 = y1 + size # 可能大于宽度
dx = max(0, -x1)
dy = max(0, -y1)
x1 = max(0, x1)
y1 = max(0, y1)
edx = max(0, x2 - width)
edy = max(0, y2 - height)
x2 = min(width, x2)
y2 = min(height, y2)
# print('box->', 'x1:', x1, 'y1:', y1, 'x2:', x2, 'y2:', y2,
# 'in', 'w:', width, 'h:', height)
# print('box->', 'top:', dy, 'bottom:', edy, 'left:', dx,
# 'right:', edx, 'in', 'w:', width, 'h:', height)
cropped = image[y1:y2, x1:x2]
if (dx > 0 or dy > 0 or edx > 0 or edy > 0):
# 当裁区域越出原图时，进行必要的黑边填充
# print('padding', dy, edy, dx, edx)
cropped = cv2.copyMakeBorder(cropped, dy, edy, dx, edx,
cv2.BORDER_CONSTANT, 0)
# 缩放到112 size
cropped = cv2.resize(cropped, (image_size, image_size))
cv2.imshow('cropped', cropped)
input = cv2.resize(cropped, (image_size, image_size))
input = cv2.cvtColor(input, cv2.COLOR_BGR2RGB)
input = input.astype(np.float32) / 256.0
input = np.expand_dims(input, 0)
feed_dict = {
images_placeholder: input,
phase_train_placeholder: False
}
cv2.rectangle(image, (x1, y1), (x2, y2), (0, 255, 0))
pre_landmarks = sess.run(landmarks, feed_dict=feed_dict)
pre_landmark = pre_landmarks[0]
pre_landmark = pre_landmark.reshape(-1, 2) * [size, size]
for (x, y) in pre_landmark.astype(np.int32):
cv2.circle(image, (x1 + x, y1 + y), 1, (0, 0, 255))
cv2.imshow('0', image)
if cv2.waitKey(10) == 27:
break
def read_images():
meta_file = './test_models2/model.meta'
ckpt_file = './test_models2/model.ckpt-59'
main_dir = './test_models2/DSM_lina/Lina15'
landmark_file = '%s/landmarks1.txt' % main_dir
result_image_path = '%s/result_image' % main_dir
if not os.path.exists(result_image_path):
os.mkdir(result_image_path)
image_size = 112
with tf.Graph().as_default():
with tf.Session(config=tf.ConfigProto(
log_device_placement=True)) as sess:
print('Loading feature extraction model.')
saver = tf.train.import_meta_graph(meta_file)
saver.restore(tf.get_default_session(), ckpt_file)
graph = tf.get_default_graph()
images_placeholder = graph.get_tensor_by_name('image_batch:0')
phase_train_placeholder = graph.get_tensor_by_name('phase_train:0')
landmark_total = graph.get_tensor_by_name(
'pfld_inference/fc/BiasAdd:0')
mtcnn = MTCNN()
sum_time = 0.
num_face_box = 0
with open(landmark_file, 'r') as fr:
lines = fr.readlines()
for index, line in enumerate(lines):
line_split = line.strip('\t\n').split(' ', 1)
image_path = '%s/%s' % (main_dir, line_split[0])
image = cv2.imread(image_path)
if image is None:
continue
height, width, _ = image.shape
boxes = mtcnn.predict(image)
# boxes, _ = predict(image)
for box in boxes:
start_time = time.time()
x1, y1, x2, y2 = (box[:4] + 0.5).astype(np.int32)
w = x2 - x1 + 1
h = y2 - y1 + 1
size = int(max([w, h]) * 1.1)
cx = x1 + w // 2
cy = y1 + h // 2
x1 = cx - size // 2
x2 = x1 + size
y1 = cy - size // 2
y2 = y1 + size
dx = max(0, -x1)
dy = max(0, -y1)
x1 = max(0, x1)
y1 = max(0, y1)
edx = max(0, x2 - width)
edy = max(0, y2 - height)
x2 = min(width, x2)
y2 = min(height, y2)
cropped = image[y1:y2, x1:x2]
if dx > 0 or dy > 0 or edx > 0 or edy > 0:
cropped = cv2.copyMakeBorder(
cropped, dy, edy, dx, edx, cv2.BORDER_CONSTANT,
0)
cropped = cv2.resize(cropped, (image_size, image_size))
input = cv2.resize(cropped, (image_size, image_size))
input = cv2.cvtColor(input, cv2.COLOR_BGR2RGB)
input = input.astype(np.float32) / 256.0
input = np.expand_dims(input, 0)
feed_dict = {
images_placeholder: input,
phase_train_placeholder: False
}
pre_landmarks = sess.run(landmark_total,
feed_dict=feed_dict)
pre_landmark = pre_landmarks[0]
pre_landmark = pre_landmark.reshape(-1,
2) * [size, size]
sum_time += (time.time() - start_time)
num_face_box += 1
cv2.rectangle(image, (x1, y1), (x2, y2), (0, 255, 0))
for (x, y) in pre_landmark.astype(np.int32):
cv2.circle(image, (x1 + x, y1 + y), 1, (0, 0, 255),
2)
# points = line_split[1].split(',')
#
# for i in range(68):
# cv2.circle(image, (int(float(points[i*2])), int(float(points[i*2+1]))), 1, (255, 0, 0), 2)
# for i in range(68):
# cv2.circle(image, (int(float(points[i*2]) * width), int(float(points[i*2+1])*height)),
# 1, (255, 0, 0), 2)
# cv2.imwrite('%s/%s' % (result_image_path, line_split[0].rsplit('/', 1)[1]), image)
cv2.imshow('0', image)
if cv2.waitKey(10) == 27:
break
# img_path = './test_models2/DSM_lina/01.bmp'
#
# image_size = 112
# with tf.Graph().as_default():
# with tf.Session() as sess:
# print('Loading feature extraction model.')
# saver = tf.train.import_meta_graph(meta_file)
# saver.restore(tf.get_default_session(), ckpt_file)
#
# graph = tf.get_default_graph()
# images_placeholder = graph.get_tensor_by_name('image_batch:0')
# phase_train_placeholder = graph.get_tensor_by_name('phase_train:0')
# landmark_total = graph.get_tensor_by_name('pfld_inference/fc/BiasAdd:0')
#
#
# image = cv2.imread(img_path)
# if image is None:
# print('Error:image is None!')
# height, width, _ = image.shape
# boxes, _ = predict(image)
# for box in boxes:
# x1, y1, x2, y2 = (box[:4] + 0.5).astype(np.int32)
# w = x2 - x1 + 1
# h = y2 - y1 + 1
#
# size = int(max([w, h]) * 1.1)
# cx = x1 + w // 2
# cy = y1 + h // 2
# x1 = cx - size // 2
# x2 = x1 + size
# y1 = cy - size // 2
# y2 = y1 + size
#
# dx = max(0, -x1)
# dy = max(0, -y1)
# x1 = max(0, x1)
# y1 = max(0, y1)
#
# edx = max(0, x2 - width)
# edy = max(0, y2 - height)
# x2 = min(width, x2)
# y2 = min(height, y2)
#
# cropped = image[y1:y2, x1:x2]
# if dx > 0 or dy > 0 or edx > 0 or edy > 0:
# cropped = cv2.copyMakeBorder(cropped, dy, edy, dx, edx, cv2.BORDER_CONSTANT, 0)
# cropped = cv2.resize(cropped, (image_size, image_size))
#
# input = cv2.resize(cropped, (image_size, image_size))
# input = cv2.cvtColor(input, cv2.COLOR_BGR2RGB)
# input = input.astype(np.float32) / 256.0
# input = np.expand_dims(input, 0)
#
# feed_dict = {
# images_placeholder: input,
# phase_train_placeholder: False
# }
# cv2.rectangle(image, (x1, y1), (x2, y2), (0, 255, 0))
# pre_landmarks = sess.run(landmark_total, feed_dict=feed_dict)
# pre_landmark = pre_landmarks[0]
#
# pre_landmark = pre_landmark.reshape(-1, 2) * [size, size]
# for (x, y) in pre_landmark.astype(np.int32):
# cv2.circle(image, (x1 + x, y1 + y), 1, (0, 0, 255), 2)
#
# cv2.imshow('0', image)
# cv2.waitKey(0)
average_time = sum_time * 1000.0 / num_face_box
print("nums:{}, sum time:{:.3f}s, avg time:{:.3f}ms\n".format(
num_face_box, sum_time, average_time))