def align_multi(cfg, image, min_confidence=0.97, limits=None):
boxes = []
landmarks = []
detector = MTCNN()
faces = detector.detect_faces(image)
refrence = get_reference_facial_points(default_square=True)
for face in faces:
if face['confidence'] < min_confidence:
continue
boxes.append(face['box'])
landmark = []
for name, points in face['keypoints'].items():
landmark.append(list(points))
landmarks.append(landmark)
if limits:
boxes = boxes[:limits]
landmarks = landmarks[:limits]
faces = []
for landmark in landmarks:
warped_face = warp_and_crop_face(image,
landmark,
reference_pts=refrence,
crop_size=(cfg['input_size'],
cfg['input_size']))
faces.append(warped_face)
return np.array(boxes), np.array(landmarks), np.array(faces)
def align_face(img_path, resize_face_size=112):
img = Image.open(img_path)
bounding_boxes, landmarks = detect_faces(img)
scale = resize_face_size / 112.
reference = get_reference_facial_points(default_square = True) * scale
facial5points = [[landmarks[0][j], landmarks[0][j + 5]] for j in range(5)]
warped_face = warp_and_crop_face(np.array(img), facial5points, reference, crop_size=(resize_face_size, resize_face_size))
return warped_face
def prepare_facebank(cfg, model):
names = ['Unknown']
embeddings = []
detector = MTCNN()
for name in os.listdir(cfg['face_bank']):
if os.path.isfile(name):
continue
else:
emb = []
for file in tqdm(os.listdir(os.path.join(cfg['face_bank'], name))):
if not os.path.isfile(
os.path.join(cfg['face_bank'], name, file)):
continue
else:
image = cv2.imread(
os.path.join(cfg['face_bank'], name, file))
image = cv2.resize(image,
(cfg['input_size'], cfg['input_size']))
face = detector.detect_faces(image)
if len(face) > 0:
face = face[0]
refrence = get_reference_facial_points(
default_square=True)
landmark = []
for _, points in face['keypoints'].items():
landmark.append(list(points))
warped_face = warp_and_crop_face(
image,
landmark,
reference_pts=refrence,
crop_size=(cfg['input_size'], cfg['input_size']))
image = np.array(warped_face)
image = image.astype(np.float32) / 255.
if len(image.shape) == 3:
image = np.expand_dims(image, 0)
emb.append(l2_norm(model(image)).numpy())
if len(emb) == 0:
continue
emb = np.array(emb)
mean = np.mean(emb, axis=0)
embeddings.append(mean)
names.append(name)
embeddings = np.array(embeddings)
names = np.array(names)
np.save(os.path.join('data', 'facebank.npy'), embeddings)
np.save(os.path.join('data', 'names.npy'), names)
return embeddings, names
def align(cfg, image):
detector = MTCNN()
face = detector.detect_faces(image)[0]
refrence = get_reference_facial_points(default_square=True)
landmark = []
for name, points in face['keypoints'].items():
landmark.append(list(points))
warped_face = warp_and_crop_face(image,
landmark,
reference_pts=refrence,
crop_size=(cfg['input_size'],
cfg['input_size']))
return warped_face
def large_pose_crop(img, lmrks, pose, reference, args):
scale_size = math.ceil(args.crop_size / 144 *
48) # According to LightCNN cropping strategy
if args.WarpAffine:
warped_face = warp_and_crop_face(np.array(img),
lmrks,
reference,
crop_size=(args.crop_size,
args.crop_size))
elif args.Align_Eyes_Crop:
if pose == '110' or pose == '120' or pose == '010' or pose == '240':
eye_middle_x = (lmrks[0][0] + lmrks[1][0]) / 2
eye_middle_y = (lmrks[0][1] + lmrks[1][1]) / 2
mouth_middle_y = (lmrks[3][1] + lmrks[4][1]) / 2
Scale_thresh = scale_size / (mouth_middle_y - eye_middle_y)
resize_img = cv2.resize(img, (int(img.shape[1] * Scale_thresh),
int(img.shape[0] * Scale_thresh)))
new_eye_mid_x, new_eye_mid_y = int(
eye_middle_x * Scale_thresh), int(eye_middle_y * Scale_thresh)
facial_boundary = facial_bound(new_eye_mid_y, new_eye_mid_x,
args.crop_size)
[upper_bound, lower_bound, left_bound,
right_bound], img = facial_pad(facial_boundary, resize_img)
else:
scale = scale_size / (((lmrks[2][0] - lmrks[3][0])**2 +
(lmrks[2][1] - lmrks[3][1])**2)**0.5)
atan = math.atan2(lmrks[0][1] - lmrks[1][1],
lmrks[0][0] - lmrks[1][0]) * 57.2957795 + 180
center = (lmrks[2][0], lmrks[2][1])
M = cv2.getRotationMatrix2D(center, atan, scale)
rotated_face = cv2.warpAffine(img, M, (img.shape[1], img.shape[0]))
facial_boundary = facial_bound(lmrks[2][1], lmrks[2][0],
args.crop_size)
[upper_bound, lower_bound, left_bound,
right_bound], img = facial_pad(facial_boundary, rotated_face)
warped_face = img[upper_bound:lower_bound, left_bound:right_bound, :]
return warped_face
for subfolder in tqdm(os.listdir(source_root)):
if not os.path.isdir(os.path.join(dest_root, subfolder)):
os.mkdir(os.path.join(dest_root, subfolder))
for image_name in os.listdir(os.path.join(source_root, subfolder)):
print("Processing\t{}".format(
os.path.join(source_root, subfolder, image_name)))
img = Image.open(os.path.join(source_root, subfolder, image_name))
try: # Handle exception
_, landmarks = detect_faces(img)
except Exception:
print("{} is discarded due to exception!".format(
os.path.join(source_root, subfolder, image_name)))
continue
if len(
landmarks
) == 0: # If the landmarks cannot be detected, the img will be discarded
print("{} is discarded due to non-detected landmarks!".format(
os.path.join(source_root, subfolder, image_name)))
continue
facial5points = [[landmarks[0][j], landmarks[0][j + 5]]
for j in range(5)]
warped_face = warp_and_crop_face(np.array(img),
facial5points,
reference,
crop_size=(crop_size, crop_size))
img_warped = Image.fromarray(warped_face)
if image_name.split('.')[-1].lower() not in ['jpg', 'jpeg'
]: #not from jpg
image_name = '.'.join(image_name.split('.')[:-1]) + '.jpg'
img_warped.save(os.path.join(dest_root, subfolder, image_name))
crop_size = args.crop_size # specify size of aligned faces, align and crop with padding
scale = crop_size / 112.
reference = get_reference_facial_points(default_square = True) * scale
cwd = os.getcwd() # delete '.DS_Store' existed in the source_root
os.chdir(source_root)
os.system("find . -name '*.DS_Store' -type f -delete")
os.chdir(cwd)
if not os.path.isdir(dest_root):
os.mkdir(dest_root)
for subfolder in tqdm(os.listdir(source_root)):
if not os.path.isdir(os.path.join(dest_root, subfolder)):
os.mkdir(os.path.join(dest_root, subfolder))
for image_name in os.listdir(os.path.join(source_root, subfolder)):
print("Processing\t{}".format(os.path.join(source_root, subfolder, image_name)))
img = Image.open(os.path.join(source_root, subfolder, image_name))
try: # Handle exception
_, landmarks = detect_faces(img)
except Exception:
print("{} is discarded due to exception!".format(os.path.join(source_root, subfolder, image_name)))
continue
if len(landmarks) == 0: # If the landmarks cannot be detected, the img will be discarded
print("{} is discarded due to non-detected landmarks!".format(os.path.join(source_root, subfolder, image_name)))
continue
facial5points = [[landmarks[0][j], landmarks[0][j + 5]] for j in range(5)]
warped_face = warp_and_crop_face(np.array(img), facial5points, reference, crop_size=(crop_size, crop_size))
img_warped = Image.fromarray(warped_face)
img_warped.save(os.path.join(dest_root, subfolder, image_name.split('.')[0] + '.jpg'))
