def gen_rnet_data(data_dir,
anno_file,
pnet_model_file,
prefix_path='',
use_cuda=True,
vis=False):
# load the pnet and pnet_detector
mtcnn_detector = MtcnnDetector(p_model_path=pnet_model_file,
r_model_path=None,
o_model_path=None,
min_face_size=12,
use_cuda=True)
device = mtcnn_detector.device
imagedb = ImageDB(anno_file, mode="test", prefix_path=prefix_path)
imdb = imagedb.load_imdb()
image_reader = TestImageLoader(imdb, 1, False)
all_boxes = []
batch_idx = 0
for databatch in image_reader:
if batch_idx % 100 == 0:
print("%d images done" % batch_idx)
im = databatch
t = time.time()
boxes, boxes_align = mtcnn_detector.detect_pnet(im)
if boxes_align is None:
all_boxes.append(np.array([]))
continue
if vis:
vision.vis_face(im, boxes_align)
t1 = time.time() - t
print('time cost for image {} / {} : {:.4f}'.format(
batch_idx, image_reader.size, t1))
all_boxes.append(boxes_align)
batch_idx += 1
save_path = config.TRAIN_DATA_DIR
if not os.path.exists(save_path):
os.mkdir(save_path)
save_file = os.path.join(save_path,
"pnet_detections_%d.pkl" % int(time.time()))
with open(save_file, 'wb') as f:
pickle.dump(all_boxes, f, pickle.HIGHEST_PROTOCOL)
save_file = '/home/liujing/Codes/MTCNN/data/pnet_detections_1532530263.pkl'
get_rnet_sample_data(data_dir, anno_file, save_file, prefix_path)
def gen_onet_data(data_dir, anno_file, pnet_model_file, rnet_model_file, prefix_path='', use_cuda=True, vis=False):
pnet, rnet, _ = create_mtcnn_net(p_model_path=pnet_model_file, r_model_path=rnet_model_file, use_cuda=use_cuda)
mtcnn_detector = MtcnnDetector(pnet=pnet, rnet=rnet, min_face_size=12)
imagedb = ImageDB(anno_file, mode="test", prefix_path=prefix_path)
imdb = imagedb.load_imdb()
image_reader = TestImageLoader(imdb,1,False)
all_boxes = list()
batch_idx = 0
for databatch in image_reader:
if batch_idx % 100 == 0:
print("%d images done" % batch_idx)
im = databatch
t = time.time()
#detect an image by pnet and rnet
p_boxes, p_boxes_align = mtcnn_detector.detect_pnet(im=im)
boxes, boxes_align = mtcnn_detector.detect_rnet(im=im, dets=p_boxes_align)
if boxes_align is None:
all_boxes.append(np.array([]))
batch_idx += 1
continue
if vis:
rgb_im = cv2.cvtColor(np.asarray(im), cv2.COLOR_BGR2RGB)
vision.vis_two(rgb_im, boxes, boxes_align)
t1 = time.time() - t
print 'time cost for image ', batch_idx, '/', image_reader.size, ': ', t1
all_boxes.append(boxes_align)
batch_idx += 1
save_path = config.MODLE_STORE_DIR
if not os.path.exists(save_path):
os.mkdir(save_path)
save_file = os.path.join(save_path, "rnet_detections_%d.pkl" % int(time.time()))
with open(save_file, 'wb') as f:
cPickle.dump(all_boxes, f, cPickle.HIGHEST_PROTOCOL)
#save_file = '/home/wujiyang/FaceProjects/MTCNN_TRAIN/model_store/rnet_detections_1527304558.pkl'
get_onet_sample_data(data_dir, anno_file, save_file, prefix_path)
Created on Sat May 26 13:57:59 2018
@author: wujiyang
"""
import sys
sys.path.append("/home/wujiyang/FaceProjects/MTCNN_TRAIN")
import cv2
from tools.detect import create_mtcnn_net, MtcnnDetector
import tools.vision as vision
if __name__ == '__main__':
pnet, rnet, onet = create_mtcnn_net(p_model_path="./model_store/pnet_model_final.pt",
r_model_path="./model_store/rnet_model_final.pt",
o_model_path="./model_store/onet_model_final.pt",
use_cuda=False)
mtcnn_detector = MtcnnDetector(pnet=pnet, rnet=rnet, onet=onet, min_face_size=24)
img = cv2.imread("./test2.jpg")
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
bboxs, landmarks = mtcnn_detector.detect_face(img)
#print bboxs.shape[0]
#print landmarks.shape[0]
vision.vis_face(img, bboxs, landmarks)
score = clf.decision_function(feature)
return pred, score
if __name__ == '__main__':
# 下面四行做何用？
parser = ap.ArgumentParser()
parser.add_argument('-m',
"--model_name",
help="model name",
default="re_train_model_V4.m")
args = vars(parser.parse_args())
model = args["model_name"]
# MTCNN人脸检测器初始化
mtcnn_detector = MtcnnDetector(min_face_size=24,
use_cuda=False) # mtcnn_detector的数据格式需要解析
logger.info("Init the MtcnnDetector.")
project_root = pathlib.Path()
model_path = project_root / "model" / str(model)
# 从摄像头读取图片，目前已不需要
cap = cv2.VideoCapture(0)
cap.set(3, 640) # 设置分辨率
cap.set(4, 480)
cap.set(15, -6.0)
while True:
ret, frame = cap.read(0) # 帧提取，ret无用
start = time.time()
bboxs, landmarks = mtcnn_detector.detect_face(
frame) # landmarks无用，bbox人脸区域坐标矩阵
detect_time = time.time() - start
def draw_images(img, bboxs, landmarks): # 在图片上绘制人脸框及特征点
num_face = bboxs.shape[0]
for i in range(num_face):
cv2.rectangle(img, (int(bboxs[i, 0]), int(bboxs[i, 1])),
(int(bboxs[i, 2]), int(bboxs[i, 3])), (0, 255, 0), 3)
for p in landmarks:
for i in range(5):
cv2.circle(img, (int(p[2 * i]), int(p[2 * i + 1])), 6, (0, 0, 255),
-1)
return img
if __name__ == '__main__':
mtcnn_detector = MtcnnDetector(min_face_size=24,
use_cuda=False) # 加载模型参数，构造检测器
logger.info("Init the MtcnnDetector.")
project_root = pathlib.Path()
inputPath = project_root / "data" / "test_images"
outputPath = project_root / "data" / "you_result"
outputPath.mkdir(exist_ok=True)
start = time.time()
for num, input_img_filename in enumerate(inputPath.iterdir()):
logger.info("Start to process No.{} image.".format(num))
img_name = input_img_filename.name
logger.info("The name of the image is {}.".format(img_name))
img = cv2.imread(str(input_img_filename))
RGB_image = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
bboxs, landmarks = mtcnn_detector.detect_face(
# r_y = min(bboxs[i, 3] + 0.4 * b_h, h)
r_y = bboxs[i, 3]
cb_0 = (0.4 * b_w) if bboxs[i, 0] > (0.4 * b_w) else bboxs[i, 0]
cb_1 = (0.4 * b_h) if bboxs[i, 1] > (0.4 * b_h) else bboxs[i, 1]
cb_2 = (1.4 * b_w) if bboxs[i, 0] > (0.4 * b_w) else bboxs[i, 0] + b_w
cb_3 = cb_2 + b_h
cropped.append(img[int(l_y):int(r_y), int(l_x):int(r_x)])
croped_bboxs.append(int(cb_0))
croped_bboxs.append(int(cb_1))
croped_bboxs.append(int(cb_2))
croped_bboxs.append(int(cb_3))
return cropped, croped_bboxs
if __name__ == '__main__':
mtcnn_detector = MtcnnDetector(min_face_size=24, use_cuda=True)
for i_path in range(len(data_path_list)):
data_path = data_path_list[i_path]
img_name_list = os.listdir(in_path + data_path)
num_imgs = len(img_name_list)
cropped_img_list = list()
cropped_bboxs_list = list()
img_list = list()
bboxs_list = list()
print("Now at Part " + str(i_path) + " of " +
str(len(data_path_list)) + " " + data_path +
" | number of imgs: " + str(num_imgs))
for i in range(num_imgs):
img_name = img_name_list[i]
print(">>> Processing: " + str(i) + " of " + str(num_imgs) + " " +
def demo(path):
#trained model
p_model_path = "./model/pnet_epoch_train.pt"
r_model_path = "./model/rnet_epoch_train.pt"
o_model_path = "./model/onet_epoch_train.pt"
pnet, rnet, onet = create_mtcnn_net(p_model_path=p_model_path,
r_model_path=r_model_path,
o_model_path=o_model_path,
use_cuda=False)
mtcnn_detector = MtcnnDetector(pnet=pnet,
rnet=rnet,
onet=onet,
min_face_size=24,
threshold=[0.6, 0.7, 0.7])
img = cv2.imread(path)
bboxs = mtcnn_detector.detect_face(img)
# print box_align
c = 0
st = 0
for i in range(bboxs.shape[0]):
b = bboxs[i, :4]
l = b[2] - b[0]
h = b[3] - b[1]
s = l * h
if st < s:
st = s
cv2.rectangle(img, (int(b[0]), int(b[1])), (int(b[2]), int(b[3])),
(0, 0, 255), 2)
image_r = img[int(b[1]) - 10:int(b[3]) + 10,
int(b[0]) - 10:int(b[2]) + 10]
# cv2.imwrite("./img/" + "wxf" + ".jpg", image_r, [int(cv2.IMWRITE_PNG_COMPRESSION), 9])
image = cv2.cvtColor(image_r, cv2.COLOR_BGR2HSV)
sumRed = 0
sumYello = 0
sumblack = 0
sumwhite = 0
sumcyan = 0
for i in range(image.shape[0]):
for j in range(image.shape[1]):
if (image[i, j, 0] <= 180 & image[i, j, 0] >= 156) & (
image[i, j, 1] >= 43
& image[i, j, 1] <= 255) & (image[i, j, 2] >=
46 & image[i, j, 2] <= 255):
sumRed = sumRed + 1
if (image[i, j, 0] <= 10 & image[i, j, 0] >= 0) & (
image[i, j, 1] >= 43
& image[i, j, 1] <= 255) & (image[i, j, 2] >=
46 & image[i, j, 2] <= 255):
sumRed = sumRed + 1
if (image[i, j, 0] <= 180 & image[i, j, 0] >=
0) & (image[i, j, 1] >= 0
& image[i, j, 1] <=
255) & (image[i, j, 2] >= 0 & image[i, j, 2] <= 46):
sumblack = sumblack + 1
if (image[i, j, 0] <= 180 & image[i, j, 0] >= 0) & (
image[i, j, 1] >= 0
& image[i, j, 1] <= 30) & (image[i, j, 2] >=
221 & image[i, j, 2] <= 255):
sumwhite = sumwhite + 1
if (image[i, j, 0] <= 255 & image[i, j, 0] >= 34) & (
image[i, j, 1] >= 43
& image[i, j, 1] <= 255) & (image[i, j, 2] >=
46 & image[i, j, 2] <= 255):
sumYello = sumYello + 1
if (image[i, j, 0] <= 99 & image[i, j, 0] >= 78) & (
image[i, j, 1] >= 43
& image[i, j, 1] <= 255) & (image[i, j, 2] >=
46 & image[i, j, 2] <= 255):
sumcyan = sumcyan + 1
all = image.shape[0] * image.shape[1]
red = sumRed / all
yello = sumYello / all
black = sumblack / all
cyan = sumcyan / all
white = sumwhite / all
print('red:', red, 'yello:', yello, 'white:', white, 'black:', black,
'cyan:', cyan)
console_handler.formatter = formatter # 也可以直接给formatter赋值
def draw_images(img, bboxs, landmarks): # 在图片上绘制人脸框及特征点
num_face = bboxs.shape[0]
for i in range(num_face):
cv2.rectangle(img, (int(bboxs[i, 0]), int(bboxs[i, 1])), (int(
bboxs[i, 2]), int(bboxs[i, 3])), (0, 255, 0), 3)
for p in landmarks:
for i in range(5):
cv2.circle(img, (int(p[2 * i]), int(p[2 * i + 1])), 6, (0, 0, 255), -1)
return img
if __name__ == '__main__':
mtcnn_detector = MtcnnDetector(min_face_size=24, use_cuda=False) # 加载模型参数，构造检测器
logger.info("Init the MtcnnDetector.")
project_root = pathlib.Path()
inputPath = project_root / "data" / "test_images"
outputPath = project_root / "data" / "you_result"
outputPath.mkdir(exist_ok=True)
start = time.time()
for num, input_img_filename in enumerate(inputPath.iterdir()):
logger.info("Start to process No.{} image.".format(num))
img_name = input_img_filename.name
logger.info("The name of the image is {}.".format(img_name))
img = cv2.imread(str(input_img_filename))
RGB_image = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
bboxs, landmarks = mtcnn_detector.detect_face(RGB_image) # 检测得到bboxs以及特征点