def image_encode(args, i, item, q_out):
oitem = [item.id]
#print('flag', item.flag)
if item.flag == 0:
fullpath = item.image_path
header = mx.recordio.IRHeader(item.flag, item.label, item.id, 0)
#print('write', item.flag, item.id, item.label)
if item.aligned:
with open(fullpath, 'rb') as fin:
img = fin.read()
s = mx.recordio.pack(header, img)
q_out.put((i, s, oitem))
else:
img = cv2.imread(fullpath, args.color)
assert item.landmark is not None
img = face_preprocess.preprocess(img,
bbox=item.bbox,
landmark=item.landmark,
image_size='%d,%d' %
(args.image_h, args.image_w))
s = mx.recordio.pack_img(header,
img,
quality=args.quality,
img_fmt=args.encoding)
q_out.put((i, s, oitem))
else:
header = mx.recordio.IRHeader(item.flag, item.label, item.id, 0)
#print('write', item.flag, item.id, item.label)
s = mx.recordio.pack(header, '')
q_out.put((i, s, oitem))
def get_all_faces(self, img):
str_image_size = "%d,%d"%(self.image_size[0], self.image_size[1])
bounding_boxes, points = detect_face.detect_face(img, self.det_minsize, self.pnet, self.rnet, self.onet, self.det_threshold, self.det_factor)
ret = []
for i in range(bounding_boxes.shape[0]):
bbox = bounding_boxes[i,0:4]
landmark = points[:, i].reshape((2,5)).T
aligned = face_preprocess.preprocess(img, bbox=bbox, landmark = landmark, image_size=str_image_size)
aligned = np.transpose(aligned, (2,0,1))
ret.append(aligned)
return ret
def cal_time_cost(embedding, detector, loop_time):
model = mx.mod.Module(symbol=embedding,
context=mx.cpu(0),
label_names=None)
model.bind(data_shapes=[('data', (1, 3, 112, 112))])
model.init_params()
read_img_time = 0
crop_time = 0
embedding_time = 0
for i in range(loop_time):
# 读取图片
start_time = time.time()
img = cv2.imread('Tom_Hanks_54745.png')
end_time = time.time()
read_img_time += end_time - start_time
#print('cost of image read:' + str(end_time - start_time))
for i in range(loop_time):
# 从图片中crop人脸
start_time = time.time()
ret = detector.detect_face(img, det_type=0)
bbox, points = ret
bbox = bbox[0, 0:4]
points = points[0, :].reshape((2, 5)).T
nimg = face_preprocess.preprocess(img,
bbox,
points,
image_size='112,112')
nimg = cv2.cvtColor(nimg, cv2.COLOR_BGR2RGB)
aligned = np.transpose(nimg, (2, 0, 1))
end_time = time.time()
crop_time += end_time - start_time
#print('cost of crop input:' + str(end_time - start_time))
for i in range(loop_time):
# 根据人脸获取embedding
start_time = time.time()
input_blob = np.expand_dims(aligned, axis=0)
data = mx.nd.array(input_blob)
db = mx.io.DataBatch(data=(data, ))
model.forward(db, is_train=False)
embedding = model.get_outputs()[0].asnumpy()
#embedding = sklearn.preprocessing.normalize(embedding).flatten()
end_time = time.time()
embedding_time += end_time - start_time
#print('cost of generate features:' + str(end_time - start_time))
return read_img_time / loop_time, crop_time / loop_time, embedding_time / loop_time
def get_input(self, face_img):
ret = self.detector.detect_face(face_img, det_type = self.args.det)
if ret is None:
return None
bbox, points = ret
if bbox.shape[0]==0:
return None
bbox = bbox[0,0:4]
points = points[0,:].reshape((2,5)).T
#print(bbox)
#print(points)
nimg = face_preprocess.preprocess(face_img, bbox, points, image_size='112,112')
nimg = cv2.cvtColor(nimg, cv2.COLOR_BGR2RGB)
aligned = np.transpose(nimg, (2,0,1))
return aligned
def face_align(src_img, fd, args):
src_image = ImageData(src_img, fd.input_shape)
src_image.resize_image(mode='pad')
dets, lmks = fd.detector.detect(src_image.transformed_image,
threshold=args.fd_conf)
if dets.shape[0] > 0:
bindex = fd.get_max_face(dets)
landmark = lmks[bindex]
bbox = dets[bindex, :4].astype(np.int)
image_size = '%d,%d' % (args.fr_input_size[1], args.fr_input_size[0])
warped = preprocess(src_image.transformed_image,
bbox,
landmark,
image_size=image_size)
return warped
else:
return None
def get_feature(self, face_img):
#face_img is bgr image
ret = self.detector.detect_face_limited(face_img,
det_type=self.args.det)
if ret is None:
return None
bbox, points = ret
if bbox.shape[0] == 0:
return None
bbox = bbox[0, 0:4]
points = points[0, :].reshape((2, 5)).T
#print(bbox)
#print(points)
nimg = face_preprocess.preprocess(face_img,
bbox,
points,
image_size='112,112')
nimg = cv2.cvtColor(nimg, cv2.COLOR_BGR2RGB)
aligned = np.transpose(nimg, (2, 0, 1))
#print(nimg.shape)
embedding = None
for flipid in [0, 1]:
if flipid == 1:
if self.args.flip == 0:
break
do_flip(aligned)
input_blob = np.expand_dims(aligned, axis=0)
data = mx.nd.array(input_blob)
db = mx.io.DataBatch(data=(data, ))
self.model.forward(db, is_train=False)
_embedding = self.model.get_outputs()[0].asnumpy()
#print(_embedding.shape)
if embedding is None:
embedding = _embedding
else:
embedding += _embedding
embedding = sklearn.preprocessing.normalize(embedding).flatten()
return embedding
def get_aligned_face(self, img, force = False):
#print('before det', img.shape)
bounding_boxes, points = detect_face.detect_face(img, self.det_minsize, self.pnet, self.rnet, self.onet, self.det_threshold, self.det_factor)
#if bounding_boxes.shape[0]==0:
# fimg = np.copy(img)
# do_flip(fimg)
# bounding_boxes, points = detect_face.detect_face(fimg, self.det_minsize, self.pnet, self.rnet, self.onet, self.det_threshold, self.det_factor)
if bounding_boxes.shape[0]==0 and force:
print('force det', img.shape)
bounding_boxes, points = detect_face.detect_face(img, self.det_minsize, self.pnet, self.rnet, self.onet, [0.3, 0.3, 0.1], self.det_factor)
#bounding_boxes, points = detect_face.detect_face_force(img, None, self.pnet, self.rnet, self.onet)
#print('after det')
if bounding_boxes.shape[0]==0:
return None
bindex = 0
nrof_faces = bounding_boxes.shape[0]
det = bounding_boxes[:,0:4]
img_size = np.asarray(img.shape)[0:2]
if nrof_faces>1:
bounding_box_size = (det[:,2]-det[:,0])*(det[:,3]-det[:,1])
img_center = img_size / 2
offsets = np.vstack([ (det[:,0]+det[:,2])/2-img_center[1], (det[:,1]+det[:,3])/2-img_center[0] ])
offset_dist_squared = np.sum(np.power(offsets,2.0),0)
bindex = np.argmax(bounding_box_size-offset_dist_squared*2.0) # some extra weight on the centering
det = bounding_boxes[:,0:4]
det = det[bindex,:]
points = points[:, bindex]
landmark = points.reshape((2,5)).T
#points need to be transpose, points = points.reshape( (5,2) ).transpose()
det = np.squeeze(det)
bb = det
points = list(points.flatten())
assert(len(points)==10)
str_image_size = "%d,%d"%(self.image_size[0], self.image_size[1])
warped = face_preprocess.preprocess(img, bbox=bb, landmark = landmark, image_size=str_image_size)
warped = np.transpose(warped, (2,0,1))
print(warped.shape)
return warped