class Detect:
def __init__(self):
self.detector = MtcnnDetector(model_folder=model, ctx=mx.cpu(0), num_worker=4, accurate_landmark=False)
def detect_face(self,image):
img = cv2.imread(image)
if img is not None:
print('detect face: read img ok,')
else:
print('detect face: read img fail')
results =self.detector.detect_face(img)
boxes=[]
key_points = []
if results is not None:
#box框
boxes=results[0]
#人脸5个关键点
points = results[1]
for i in results[0]:
faceKeyPoint = []
for p in points:
for i in range(5):
faceKeyPoint.append([p[i], p[i + 5]])
key_points.append(faceKeyPoint)
else:
print('detect face: result is None')
return {"boxes":boxes,"face_key_point":key_points}
def __init__(self):
self.detector = MtcnnDetector(model_folder=model,
ctx=mx.cpu(0),
num_worker=4,
accurate_landmark=False)
def main(args):
is_test = True if args.is_test == '1' else False
_t = Timer()
detector_cast = MtcnnDetector(model_folder='./mtcnn/model',
minsize=20,
threshold=[0.1, 0.5, 0.9],
factor=0.709,
ctx=mx.gpu(args.gpu),
num_worker=4,
accurate_landmark=False)
detector_candi = MtcnnDetector(model_folder='./mtcnn/model',
minsize=20,
threshold=[0.5, 0.5, 0.9],
factor=0.709,
ctx=mx.gpu(args.gpu),
num_worker=4,
accurate_landmark=False)
embedding = FaceModel(model='./arcface/model/model-r50-am-lfw',
ctx=mx.gpu(args.gpu))
if is_test:
this_dir, json_path, save_name = osp.join(test_root, 'test'), osp.join(
test_root, 'test.json'), 'face_em_test.pkl'
else:
this_dir, json_path, save_name = osp.join(
trainval_root, 'val'), osp.join(trainval_root,
'val.json'), 'face_em_val.pkl'
data_raw = load_json(json_path)
movie_num, movie_cnt = len(data_raw.keys()), 0
face_dict = {}
# det/extract val face feat
for movie, info in data_raw.items():
face_dict.update({movie: {'cast': [], 'candidates': []}})
movie_cnt += 1
casts, casts_num = info['cast'], len(info['cast'])
candidates, candidates_num = info['candidates'], len(
info['candidates'])
for i, cast in enumerate(casts):
img_path = osp.join(this_dir, cast['img'])
img = cv2.imread(img_path)
cast_id = cast['id']
_t.tic()
fbbox, landmark = face_det_cast(img, detector_cast)
assert fbbox is not None, 'Cast: No face detected !'
ffeat = face_exfeat(img, fbbox, landmark, embedding)
_t.toc()
print('%s %d/%d ... %s %d/%d ... time: %.3f s average: %.3f s' %
(movie, movie_cnt, movie_num, cast_id, i + 1, casts_num,
_t.diff, _t.average_time))
face_dict[movie]['cast'].append({
'id': cast_id,
'fbbox': fbbox,
'ffeat': ffeat
})
for i, candidate in enumerate(candidates):
img_path = osp.join(this_dir, candidate['img'])
img = cv2.imread(img_path)
candidate_id = candidate['id']
rect = candidate['bbox']
_t.tic()
crop, fbbox, landmark = face_det_candi(img, rect, detector_candi)
if fbbox is None:
_t.toc()
print(
'%s %d/%d ... %s %d/%d ... time: %.3f s average: %.3f s' %
(movie, movie_cnt, movie_num, candidate_id, i + 1,
candidates_num, _t.diff, _t.average_time))
face_dict[movie]['candidates'].append({
'id': candidate_id,
'fbbox': None,
'ffeat': None
})
continue
ffeat = face_exfeat(crop, fbbox, landmark, embedding)
_t.toc()
print('%s %d/%d ... %s %d/%d ... time: %.3f s average: %.3f s' %
(movie, movie_cnt, movie_num, candidate_id, i + 1,
candidates_num, _t.diff, _t.average_time))
face_dict[movie]['candidates'].append({
'id': candidate_id,
'fbbox': fbbox,
'ffeat': ffeat
})
my_pickle(face_dict, osp.join('./features', save_name))