def get_feature(image_path, bbox, landmark, nets, image_shape, use_align, aligned, use_mean):
img = face_preprocess.read_image(image_path, mode='rgb')
#print(img.shape)
if img is None:
print('parse image',image_path,'error')
return None
if not aligned:
_landmark = landmark
if not use_align:
_landmark = None
#cv2.imwrite("./align/origin_%s"%image_path.split('/')[-1], img)
img = face_preprocess.preprocess(img, bbox=bbox, landmark=_landmark, image_size='%d,%d'%(image_shape[1], image_shape[2]))
else:
assert img.shape==(image_shape[1],image_shape[2],image_shape[0])
#print('already aligned', image_path, img.shape)
#img = cv2.resize(img, (image_shape[2], image_shape[1]))
#cv2.imwrite("./align/%s"%image_path.split('/')[-1], img)
if use_mean>0:
v_mean = np.array([127.5,127.5,127.5], dtype=np.float32).reshape( (1,1,3) )
img = img.astype(np.float32) - v_mean
img *= 0.0078125
img = np.transpose( img, (2,0,1) )
F = None
for net in nets:
embedding = None
#ppatch = net.patch
for flipid in [0,1]:
_img = np.copy(img)
if flipid==1:
do_flip(_img)
#nimg = np.zeros(_img.shape, dtype=np.float32)
#nimg[:,ppatch[1]:ppatch[3],ppatch[0]:ppatch[2]] = _img[:, ppatch[1]:ppatch[3], ppatch[0]:ppatch[2]]
#_img = nimg
input_blob = np.expand_dims(_img, axis=0)
data = mx.nd.array(input_blob)
db = mx.io.DataBatch(data=(data,))
net.model.forward(db, is_train=False)
_embedding = net.model.get_outputs()[0].asnumpy().flatten()
#print(_embedding.shape)
if embedding is None:
embedding = _embedding
else:
embedding += _embedding
_norm=np.linalg.norm(embedding)
embedding /= _norm
if F is None:
F = embedding
else:
#F += embedding
F = np.concatenate((F,embedding), axis=0)
_norm=np.linalg.norm(F)
F /= _norm
#print(F.shape)
return F
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 xrange(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 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 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))
input_blob = np.expand_dims(aligned, axis=0)
data = mx.nd.array(input_blob)
db = mx.io.DataBatch(data=(data,))
return db
def image_encode(args, i, item, q_out):
if item.flag == 0:
fullpath = item.image_path
if item.aligned:
with open(fullpath, 'rb') as fin:
img = fin.read()
q_out.put((i, img, fullpath))
else:
img = cv2.imread(fullpath)
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))
q_out.put((i, img, fullpath))
else:
pass
def detect_face_and_get_embedding(self, img):
thresh = 0.8
flip = False
scales = self.get_scales(img)
bboxes, rs = self.face_detector.detect(img, thresh, scales=scales, do_flip=flip)
# print('bbox:', bboxes)
if len(bboxes) <= 0:
return None, None
# print('landmark:', rs)
if rs is not None:
points = rs.astype(np.int32)
# point = points[0, :].reshape((2, 5)).T
nimg = preprocess(img, bboxes[0], points[0], image_size='112,112')
# nimg = cv2.cvtColor(nimg, cv2.COLOR_BGR2RGB)
e = np.transpose(nimg, (2, 0, 1))
embeddings = self.face_recognition.get_feature(e)
return embeddings, nimg
return None, None
def get_input_train(self, face_img):
ret = self.detector.detect_face(face_img, det_type=self.loder[0])
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 get_img(self, face_img):
#face_img is bgr image
dets, points_all = self.get_face_dets_points(face_img)
if not dets:
return None, None
# print dets
#print points_all
aligned_imgs = []
for i, bbox in enumerate(dets):
points = points_all[i]
nimg = face_preprocess.preprocess(face_img,
bbox,
points,
image_size='112,112')
#nimg = cv2.cvtColor(nimg, cv2.COLOR_BGR2RGB)
#nimg = np.transpose(nimg, (2,0,1))
aligned_imgs.append(nimg)
return aligned_imgs, dets
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)
img = cv2.imread(fullpath, args.color)
det = item.bbox
margin = 44
bb = np.zeros(4, dtype=np.int32)
bb[0] = np.maximum(det[0] - margin / 2, 0)
bb[1] = np.maximum(det[1] - margin / 2, 0)
bb[2] = np.minimum(det[2] + margin / 2, img.shape[1])
bb[3] = np.minimum(det[3] + margin / 2, img.shape[0])
img = img[bb[1]:bb[3], bb[0]:bb[2], :]
img = cv2.resize(img, (224, 224))
s = mx.recordio.pack_img(header,
img,
quality=args.quality,
img_fmt=args.encoding)
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, b'')
q_out.put((i, s, oitem))
def get_feature(self, face_img, tp, fname):
#face_img is bgr image
ret = self.detector.detect_face(face_img)
# 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)
#face_img BGR
nimg = face_preprocess.preprocess(face_img,
bbox,
points,
image_size='112,112')
# save image
# fPath = os.path.join(IMG_OUTPUT, tp, fname + ".jpg")
# cv2.imwrite(fPath, nimg)
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_input_multi(self, face_img):
img = face_img
im_shape = img.shape
scales = [1024, 1980]
target_size = scales[0]
max_size = scales[1]
im_size_min = np.min(im_shape[0:2])
im_size_max = np.max(im_shape[0:2])
#im_scale = 1.0
#if im_size_min>target_size or im_size_max>max_size:
im_scale = float(target_size) / float(im_size_min)
# prevent bigger axis from being more than max_size:
if np.round(im_scale * im_size_max) > max_size:
im_scale = float(max_size) / float(im_size_max)
print('im_scale', im_scale)
scales = [im_scale]
flip = False
ret = self.detector.detect(img, thresh, scales=scales, do_flip=False)
if ret is None:
return None
bbox, points = ret
if bbox.shape[0] == 0:
return None
bbox_list = []
aligned_list = []
for i in range(bbox.shape[0]):
bbox_ = bbox[i, 0:4]
points_ = points[i, :, :]
#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))
bbox_list.append([int(h) for h in bbox_])
aligned_list.append(aligned)
return aligned_list, bbox_list
def getGalleryFeature(detector,fr_model):
gallery_paths = glob.glob('./gallery/*')
names=[]
imgs_feats=[]
for p in gallery_paths:
if os.path.isfile(p):
names.append(p.split('/')[len(p.split('/'))-1][:-4])
img=cv2.imread(p)
results=detector.detect(img)
landmark=results[0][1]
img=preprocess(img,landmark)
img=cv2.cvtColor(img,cv2.COLOR_BGR2RGB)
img = transform(Image.fromarray(img)).unsqueeze(0).cuda()
feat=fr_model(img)
imgs_feats.append(feat)
return names,imgs_feats
def get_input_loc(self, face_img):
ret = self.detector.detect_face(face_img, det_type=self.det)
if ret is None:
return None, None, None
bbox, points = ret
if bbox.shape[0] == 0:
return None, None, None
bbox = bbox[0, 0:4]
points = points[0, :].reshape((2, 5)).T
nimg, bb = face_preprocess.preprocess(face_img,
bbox,
points,
image_size='112,112')
loc = face_img.copy()
cv2.rectangle(loc, (bb[0], bb[1]), (bb[2], bb[3]), (0, 0, 255), 2)
aligned = cv2.cvtColor(nimg, cv2.COLOR_BGR2RGB)
aligned = np.transpose(aligned, (2, 0, 1))
return aligned, nimg, loc
def despoof(self, img_bgr):
img = cv2.cvtColor(img_bgr, cv2.COLOR_BGR2RGB)
bounding_boxes, points = detect_face.detect_face(
img, self.MTCNN_params['minsize'], self.pnet, self.rnet, self.onet, self.MTCNN_params['threshold'], self.MTCNN_params['factor'])
nrof_faces = bounding_boxes.shape[0]
face_detected_flag = False
if nrof_faces >= 1:
for index, bbox in enumerate(bounding_boxes):
score = bbox[-1]
bbox = bbox[0: 4]
landmark = points[:, index].reshape((2, 5)).T
if score > 0.8:
face_detected_flag = True
# print(bbox[3] - bbox[1], bbox[2] - bbox[0])
warped = face_preprocess.preprocess(
img, bbox=bbox, landmark=landmark, image_size=(112, 112))
warped__ = preprocess_input(warped)
to_test = np.expand_dims(warped__, 0)
prediction = self.antispoofing_model.predict(to_test)
# 0.001
if prediction <= self.threshold:
# draw_box(img_bgr, [bbox[0], bbox[1], bbox[2],
# bbox[3]], color=[127, 255, 0])
# video_writer.write(img_bgr.astype(np.uint8))
self.pop_queue(1)
else:
# draw_box(img_bgr, [bbox[0], bbox[1], bbox[2],
# bbox[3]], color=[0, 0, 255])
self.pop_queue(0)
else:
self.pop_queue(0)
else:
self.pop_queue(0)
if face_detected_flag:
if self.get_result():
draw_box(img_bgr, [bbox[0], bbox[1], bbox[2],
bbox[3]], color=[127, 255, 0])
else:
draw_box(img_bgr, [bbox[0], bbox[1], bbox[2],
bbox[3]], color=[0, 0, 255])
return img_bgr
def get_input(self, face_img):
ret = self.detector.detect_face(face_img, det_type = self.args.det)
if ret is None:
return None,None
bboxs, points = ret
if bboxs.shape[0]==0:
return None,None
total = bboxs.shape[0]
aligned_imgs = []
for i in range(total):
bbox = bboxs[i,0:4]
point = points[i,:].reshape((2,5)).T
#print(bbox)
#print(points)
nimg = face_preprocess.preprocess(face_img, bbox, point, image_size='112,112')
nimg = cv2.cvtColor(nimg, cv2.COLOR_BGR2RGB)
aligned = np.transpose(nimg, (2,0,1))
aligned_imgs.append(aligned)
return aligned_imgs,bboxs
def get_input(self, face_img):
ret = self.detector.detect_face(face_img, det_type=self.args.det)
if ret is None:
return np.transpose(
cv2.cvtColor(cv2.resize(face_img, (112, 112)),
cv2.COLOR_BGR2RGB), (2, 0, 1))
bbox, points = ret
if bbox.shape[0] == 0:
return np.transpose(
cv2.cvtColor(cv2.resize(face_img, (112, 112)),
cv2.COLOR_BGR2RGB), (2, 0, 1))
bbox = bbox[0, 0:4]
points = points[0, :].reshape((2, 5)).T
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 get_input(self, face_img):
ret = self.detector.detect_face(face_img)
if ret is None:
print('This image no faces..')
return None
bbox, points = ret
#print(bbox)
if bbox.shape[0] == 0:
print('This image no faces..')
return None
bbox = bbox[0, 0:4]
points = points[0, :].reshape((2, 5)).T # matrix tranpose
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 get_input(self, face_img, output_filename_n):
ret = self.detector.detect_face(face_img, det_type=self.args.det)
if ret is not None:
bbox, points = ret
if bbox.shape[0] != 0:
bbox = bbox[0, 0:4]
points = points[0, :].reshape((2, 5)).T
# print('bbox',bbox)
# print('points',points)
nimg = face_preprocess.preprocess(face_img,
bbox,
points,
image_size=self.image_size)
nimg = cv2.cvtColor(nimg, cv2.COLOR_BGR2RGB)
misc.imsave(output_filename_n, nimg)
return 1
else:
return 0
else:
return 0
def get_all_input(self, face_img, save_img=False):
face_num = self.max_face_number
ret = self.detector.detect_face(face_img, det_type=self.args.det)
if ret is None:
return []
bbox, points = ret
sorted_index = bbox[:, 0].argsort()
bbox = bbox[sorted_index]
points = points[sorted_index]
# print(bbox)
# print(points)
if bbox.shape[0] == 0:
return None
aligned = []
for index in range(0, len(bbox[:face_num])):
item_bbox = bbox[index, 0:4]
item_points = points[index, :].reshape((2, 5)).T
# print(bbox)
# print(points)
nimg = face_preprocess.preprocess(face_img,
item_bbox,
item_points,
image_size='112,112')
if save_img:
cv2.imwrite(
'./Temp/{}-{}.jpg'.format(time.time(), self.face_counter),
nimg)
# print(self.face_counter)
self.face_counter += 1
nimg = cv2.cvtColor(nimg, cv2.COLOR_BGR2RGB)
aligned.append(np.transpose(nimg, (2, 0, 1)))
# print(aligned)
return zip(aligned, bbox)
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:
if args.size == 0:
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)
h, w = img.shape[:2]
h_off, w_off = (h - args.size) / 2, (w - args.size) / 2
img = img[h_off:h_off + args.size, w_off:w_off + args.size, :]
s = mx.recordio.pack_img(header,
img,
quality=args.quality,
img_fmt=args.encoding)
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 detect_face_and_get_embedding_test_2(self, img):
thresh = 0.8
flip = False
scales = self.get_scales(img)
bboxes, rs = self.face_detector.detect(img, thresh, scales=scales, do_flip=flip)
if len(bboxes) <= 0:
return None, None
embeddings = []
bbox_list = []
if rs is not None:
points = rs.astype(np.int32)
for i, bbox in enumerate(bboxes):
point = points[i, :].reshape((2, 5)).T
nimg = preprocess(img, bbox, point, image_size='112,112')
nimg = cv2.cvtColor(nimg, cv2.COLOR_BGR2RGB)
x = np.transpose(nimg, (2, 0, 1))
embedding = self.face_recognition.get_feature(x)
embeddings.append(embedding)
bbox_list.append(bbox)
return embeddings, bbox_list
return None, None
def get_input(self, face_img):
print('Input shape: {}'.format(face_img.shape))
ret = self.detector.detect_face(face_img, det_type = self.args.det)
if ret is None:
return None
# print('MTCNN return: {}'.format(ret))
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)
# cv2.imshow('window', nimg)
# cv2.waitKey(0)
aligned = np.transpose(nimg, (2,0,1))
return aligned
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
print("Detected box", bbox.shape, points.shape)
aligned_faces = []
for k, bb in enumerate(bbox):
bb = bb[:4]
pp = points[k].reshape((2, 5)).T
nimg = face_preprocess.preprocess(face_img,
bb,
pp,
image_size='112,112')
nimg = cv2.cvtColor(nimg, cv2.COLOR_BGR2RGB)
aligned_faces.append(np.transpose(nimg, (2, 0, 1)))
aligned_faces = np.array(aligned_faces)
print("Aligned bbox", aligned_faces.shape)
return aligned_faces, bbox, points
def mtcnn_align(img):
ret = self.detector.detect_face_limited(face_img,
det_type=self.args.det)
if ret is None:
# detected = False
bbox, points = None, None
else:
bbox, points = ret
if bbox.shape[0] == 0:
# detected = False
bbox, points = None, None
else:
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')
return nimg
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]
print(bbox)
# cv2.imshow("face", face_img[int(bbox[1]): int(bbox[3]), int(bbox[0]): int(bbox[2])])
# cv2.waitKey(0)
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 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 xrange(bounding_boxes.shape[0]):
bbox = bounding_boxes[i, 0:4]
# print(bbox)
cv2.rectangle(img, (int(bbox[0]), int(bbox[1])),
(int(bbox[2]), int(bbox[3])), (0, 0, 255), 2)
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)
img = cv2.resize(img, (int(img.shape[1] / 4), int(img.shape[0] / 4)))
cv2.imshow("faces", img)
cv2.waitKey(0)
return ret
def image_encode(args, i, item, q_out):
oitem = [item.id]
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='112,112')
s = mx.recordio.pack_img(header, img, quality=args.quality, img_fmt=args.encoding)
q_out.put((i, s, oitem))
else: #flag==1 or 2
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 image_encode(args, i, item, q_out):
oitem = [item.id]
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: #flag==1 or 2
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_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
print(len(bbox))
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)
cv2.imshow('image', nimg)
cv2.waitKey(0)
cv2.destroyAllWindows()
aligned = np.transpose(nimg, (2, 0, 1))
return aligned
def get_input1(self, face_img):
start_time = time.time()
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(face_img.shape)
print(bbox)
print(points)
nimg = face_preprocess.preprocess(face_img, bbox, points, image_size='112,112')
#nimg = face_img
nimg = cv2.cvtColor(nimg, cv2.COLOR_BGR2RGB)
print(nimg.shape)
aligned = np.transpose(nimg, (2,0,1))
print(aligned.shape)
end_time = time.time()
print('cost of get input:' + str(end_time - start_time))
return aligned
def find_faces(self, img, conf):
faces = []
boundb = []
try:
bounding_boxes, points = self.detector.detect_face(img, det_type=0)
except:
return boundb, faces
for bb, point in zip(bounding_boxes, points):
bb = bb[0:4]
bb = [int(i / conf.resize_ratio) for i in bb]
point = np.array([int(i / conf.resize_ratio) for i in point])
point = point.reshape((2, 5)).T
nimg = face_preprocess.preprocess(img,
bb,
point,
image_size='112,112')
nimg = cv2.cvtColor(nimg, cv2.COLOR_BGR2RGB)
aligned = np.transpose(nimg, (2, 0, 1))
faces.append(aligned)
boundb.append(bb)
return boundb, faces
def get_aligned_face(self, img, force=False):
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 and force:
bounding_boxes, points = detect_face.detect_face_force(
img, None, self.pnet, self.rnet, self.onet)
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
def get_feature(buffer):
global emb_size
if use_flip:
input_blob = np.zeros(
(len(buffer) * 2, 3, image_shape[1], image_shape[2]))
else:
input_blob = np.zeros((len(buffer), 3, image_shape[1], image_shape[2]))
idx = 0
for item in buffer:
img = face_preprocess.read_image(item[0], mode='rgb')
img = face_preprocess.preprocess(img,
bbox=None,
landmark=item[1],
image_size='%d,%d' %
(image_shape[1], image_shape[2]))
img = np.transpose(img, (2, 0, 1))
attempts = [0, 1] if use_flip else [0]
for flipid in attempts:
_img = np.copy(img)
if flipid == 1:
do_flip(_img)
input_blob[idx] = _img
idx += 1
data = mx.nd.array(input_blob)
db = mx.io.DataBatch(data=(data, ))
net.model.forward(db, is_train=False)
_embedding = net.model.get_outputs()[0].asnumpy()
if emb_size == 0:
emb_size = _embedding.shape[1]
print('set emb_size to ', emb_size)
embedding = np.zeros((len(buffer), emb_size), dtype=np.float32)
if use_flip:
embedding1 = _embedding[0::2]
embedding2 = _embedding[1::2]
embedding = embedding1 + embedding2
else:
embedding = _embedding
embedding = sklearn.preprocessing.normalize(embedding)
return embedding
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
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))
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()
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
def main(args):
#facenet.store_revision_info(src_path, output_dir, ' '.join(sys.argv))
dataset = face_image.get_dataset('lfw', args.input_dir)
print('dataset size', 'lfw', len(dataset))
print('Creating networks and loading parameters')
with tf.Graph().as_default():
#gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=args.gpu_memory_fraction)
#sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options, log_device_placement=False))
sess = tf.Session()
with sess.as_default():
pnet, rnet, onet = detect_face.create_mtcnn(sess, None)
minsize = 20
threshold = [0.6,0.7,0.9]
factor = 0.85
# Add a random key to the filename to allow alignment using multiple processes
#random_key = np.random.randint(0, high=99999)
#bounding_boxes_filename = os.path.join(output_dir, 'bounding_boxes_%05d.txt' % random_key)
#output_filename = os.path.join(output_dir, 'faceinsight_align_%s.lst' % args.name)
if not os.path.exists(args.output_dir):
os.makedirs(args.output_dir)
output_filename = os.path.join(args.output_dir, 'lst')
with open(output_filename, "w") as text_file:
nrof_images_total = 0
nrof = np.zeros( (5,), dtype=np.int32)
for fimage in dataset:
if nrof_images_total%100==0:
print("Processing %d, (%s)" % (nrof_images_total, nrof))
nrof_images_total += 1
#if nrof_images_total<950000:
# continue
image_path = fimage.image_path
if not os.path.exists(image_path):
print('image not found (%s)'%image_path)
continue
filename = os.path.splitext(os.path.split(image_path)[1])[0]
#print(image_path)
try:
img = misc.imread(image_path)
except (IOError, ValueError, IndexError) as e:
errorMessage = '{}: {}'.format(image_path, e)
print(errorMessage)
else:
if img.ndim<2:
print('Unable to align "%s", img dim error' % image_path)
#text_file.write('%s\n' % (output_filename))
continue
if img.ndim == 2:
img = to_rgb(img)
img = img[:,:,0:3]
_paths = fimage.image_path.split('/')
a,b = _paths[-2], _paths[-1]
target_dir = os.path.join(args.output_dir, a)
if not os.path.exists(target_dir):
os.makedirs(target_dir)
target_file = os.path.join(target_dir, b)
_minsize = minsize
_bbox = None
_landmark = None
bounding_boxes, points = detect_face.detect_face(img, _minsize, pnet, rnet, onet, threshold, factor)
nrof_faces = bounding_boxes.shape[0]
if nrof_faces>0:
det = bounding_boxes[:,0:4]
img_size = np.asarray(img.shape)[0:2]
bindex = 0
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
_bbox = bounding_boxes[bindex, 0:4]
_landmark = points[:, bindex].reshape( (2,5) ).T
nrof[0]+=1
else:
nrof[1]+=1
warped = face_preprocess.preprocess(img, bbox=_bbox, landmark = _landmark, image_size=args.image_size)
bgr = warped[...,::-1]
#print(bgr.shape)
cv2.imwrite(target_file, bgr)
oline = '%d\t%s\t%d\n' % (1,target_file, int(fimage.classname))
text_file.write(oline)
def main(args):
output_dir = os.path.expanduser(args.output_dir)
if not os.path.exists(output_dir):
os.makedirs(output_dir)
datamap = {}
pp = 0
datasize = 0
verr = 0
for line in open(args.input_dir+"_clean_list.txt", 'r'):
pp+=1
if pp%10000==0:
print('loading list', pp)
line = line.strip()[2:]
if not line.startswith('m.'):
continue
vec = line.split('/')
assert len(vec)==2
#print(line)
person = vec[0]
img = vec[1]
try:
img_id = int(img.split('.')[0])
except ValueError:
#print('value error', line)
verr+=1
continue
if not person in datamap:
labelid = len(datamap)
datamap[person] = [labelid, {img_id : 1}]
else:
datamap[person][1][img_id] = 1
datasize+=1
print('dataset size', args.name, datasize)
print('dataset err', verr)
print('Creating networks and loading parameters')
with tf.Graph().as_default():
#gpu_options = tf.GPUOptions(per_process_gpu_memory_fraction=args.gpu_memory_fraction)
#sess = tf.Session(config=tf.ConfigProto(gpu_options=gpu_options, log_device_placement=False))
sess = tf.Session()
with sess.as_default():
pnet, rnet, onet = detect_face.create_mtcnn(sess, None)
minsize = 100 # minimum size of face
threshold = [ 0.6, 0.7, 0.7 ] # three steps's threshold
factor = 0.709 # scale factor
print(minsize)
print(threshold)
print(factor)
# Add a random key to the filename to allow alignment using multiple processes
#random_key = np.random.randint(0, high=99999)
#bounding_boxes_filename = os.path.join(output_dir, 'bounding_boxes_%05d.txt' % random_key)
output_filename = os.path.join(output_dir, 'faceinsight_align_%s.lst' % args.name)
with open(output_filename, "w") as text_file:
nrof_images_total = 0
nrof_successfully_aligned = 0
nrof_changed = 0
nrof_iou3 = 0
nrof_force = 0
for line in open(args.input_dir, 'r'):
vec = line.strip().split()
person = vec[0]
img_id = int(vec[1])
v = datamap.get(person, None)
if v is None:
continue
if not img_id in v[1]:
continue
labelid = v[0]
img_str = base64.b64decode(vec[-1])
nparr = np.fromstring(img_str, np.uint8)
img = cv2.imdecode(nparr, cv2.CV_LOAD_IMAGE_COLOR)
img = img[...,::-1] #to rgb
if nrof_images_total%100==0:
print("Processing %d, (%d)" % (nrof_images_total, nrof_successfully_aligned))
nrof_images_total += 1
target_dir = os.path.join(output_dir, person)
if not os.path.exists(target_dir):
os.makedirs(target_dir)
target_path = os.path.join(target_dir, "%d.jpg"%img_id)
_minsize = minsize
fimage = edict()
fimage.bbox = None
fimage.image_path = target_path
fimage.classname = str(labelid)
if fimage.bbox is not None:
_bb = fimage.bbox
_minsize = min( [_bb[2]-_bb[0], _bb[3]-_bb[1], img.shape[0]//2, img.shape[1]//2] )
bounding_boxes, points = detect_face.detect_face(img, _minsize, pnet, rnet, onet, threshold, factor)
bindex = -1
nrof_faces = bounding_boxes.shape[0]
if fimage.bbox is None and nrof_faces>0:
det = bounding_boxes[:,0:4]
img_size = np.asarray(img.shape)[0:2]
bindex = 0
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
if fimage.bbox is not None:
if nrof_faces>0:
assert(bounding_boxes.shape[0]==points.shape[1])
det = bounding_boxes[:,0:4]
img_size = np.asarray(img.shape)[0:2]
index2 = [0.0, 0]
for i in xrange(det.shape[0]):
_det = det[i]
iou = IOU(fimage.bbox, _det)
if iou>index2[0]:
index2[0] = iou
index2[1] = i
if index2[0]>-0.3:
bindex = index2[1]
nrof_iou3+=1
if bindex<0:
bounding_boxes, points = detect_face.detect_face_force(img, fimage.bbox, pnet, rnet, onet)
bindex = 0
nrof_force+=1
if bindex>=0:
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)
warped = face_preprocess.preprocess(img, bbox=bb, landmark = landmark, image_size=args.image_size)
misc.imsave(target_path, warped)
nrof_successfully_aligned += 1
oline = '%d\t%s\t%d' % (1,fimage.image_path, int(fimage.classname))
#oline = '%d\t%s\t%d\t%d\t%d\t%d\t%d\t' % (0,fimage.image_path, int(fimage.classname), bb[0], bb[1], bb[2], bb[3])
#oline += '\t'.join([str(x) for x in points])
text_file.write("%s\n"%oline)
print('Total number of images: %d' % nrof_images_total)
print('Number of successfully aligned images: %d' % nrof_successfully_aligned)
print('Number of changed: %d' % nrof_changed)
print('Number of iou3: %d' % nrof_iou3)
print('Number of force: %d' % nrof_force)
bounding_boxes, points = detect_face.detect_face_force(img, minsize, pnet, rnet, onet)
nrof_faces = bounding_boxes.shape[0]
if nrof_faces>0:
nrof[1]+=1
else:
nrof[2]+=1
if nrof_faces>0:
det = bounding_boxes[:,0:4]
img_size = np.asarray(img.shape)[0:2]
bindex = 0
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
_bbox = bounding_boxes[bindex, 0:4]
_landmark = points[:, bindex].reshape( (2,5) ).T
warped = face_preprocess.preprocess(img, bbox=_bbox, landmark = _landmark, image_size=args.image_size)
warped = warped[...,::-1] #to bgr
_, s = cv2.imencode('.jpg', warped)
bins.append(s)
print(nrof)
outname = os.path.join(args.output, part[1]+'.bin')
with open(outname, 'wb') as f:
pickle.dump((bins, issame_list), f, protocol=pickle.HIGHEST_PROTOCOL)