count = count + 1
print("====================================")
print("img 번호: ", count)
img_frame = cv.resize(img_frame, dsize=(0, 0), fx=1, fy=1)
# if wanted == "June":
# matrix=cv.getRotationMatrix2D( (width / 2 , height / 2) , 90 , 1 )
# img_frame=cv.warpAffine( img_frame , matrix , (width , height) )
# img_frame = cv.flip(img_frame, 0) #상하반전
gray = cv.cvtColor(img_frame, cv.COLOR_BGR2GRAY)
dets = detector(gray, 1)
print("얼굴 갯수:", "{}개".format(len(dets)))
for face in dets:
fa = FaceAligner(predictor,
desiredLeftEye=(0.3, 0.3),
desiredFaceWidth=112)
faceAligned = fa.align(img_frame, gray, face.rect)
cut = copy.deepcopy(faceAligned)
x = face.rect.left()
y = face.rect.top()
w = face.rect.right() - x
h = face.rect.bottom() - y #bounding box가 작아서 shape 끝점으로 대체
cv.rectangle(img_frame, (x, y), (x + w, y + h), (0, 0, 255), 2)
print("x:", x, "y:", y, "w:", w, "h:", h)
#faceAligned = cv.resize(faceAligned,dsize = (112,112), fx=1,fy=1)
cv.imwrite('face.jpg', faceAligned)
faceAligned = Image.open('face.jpg')
# faceAligned = Image.fromarray(faceAligned)
faceAligned = transforms(faceAligned).unsqueeze(0)
from TinyFacesDetector import TinyFacesDetector
import dlib
from Utils import Utils
from FaceAligner import FaceAligner
faces_out_folder = "./output/"
image_path="sample.jpg"
model_pkl="weights.pkl"
Utils.mkdir_if_not_exist(faces_out_folder)
tiny_faces_detector = TinyFacesDetector(model_pkl,use_gpu=True)
#init the face landmarks detector
predictor_5_face_landmarks = dlib.shape_predictor("shape_predictor_5_face_landmarks.dat")
# tight face aligner : padding = 0.2
aligner_tight = FaceAligner(face_size=112, face_padding=0.2, predictor_5_face_landmarks=predictor_5_face_landmarks)
face_rects=tiny_faces_detector.detect(image_path,nms_thresh=0.1,prob_thresh=0.5,min_conf=0.9)
face_indx=0
for rect in face_rects:
face_indx+=1
aligner_tight.out_dir= faces_out_folder
aligner_tight.align_face(image_path,rect,str(face_indx)+'.jpg')
print("DDDDDDDDDDDD")
image = cv2.imread(args.img_path)
path = '/home/daehyeon/DepthNets/pipeline'
# image = cv2.resize(image, None, fx=1, fy=1, interpolation=cv2.INTER_AREA)
# Create a HOG face detector using the built-in dlib class
# Load the image into an array
start = time.time()
try:
faces_cnn = face_detector(image, 1)
except:
pass
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
# kdkd 원래는 0.375 였음
for face in faces_cnn:
fa = FaceAligner(predictor,
desiredLeftEye=(0.3, 0.3),
desiredFaceWidth=256)
faceAligned = fa.align(image, gray, face.rect)
cv2.imwrite(path + '/face_alignmented/{}.png'.format("source"),
faceAligned)
# cv2.imshow("Aligned", faceAligned)
end = time.time()
cv2.waitKey()
cv2.destroyAllWindows()
break
# dir(train_data.root)
# train_data = torchvision.datasets.ImageFolder(root='/home/daehyeon/hdd/deepfake_1st/fake/',)
# count = 0
# path = '/home/daehyeon/DepthNets/pipeline'
weight='./mmod_human_face_detector.dat'
face_detector=dlib.cnn_face_detection_model_v1( weight )
ALL=list( range( 0 , 5 ) )
dir = '/home/daehyeon/hdd/deepfake_1st/fake/'
train_data = torchvision.datasets.ImageFolder(root=dir,) # 혹은 os.list_dir(dir)로도 data list 설정가능
count = 0
for i in range(len(train_data.imgs)):
path = train_data.imgs[i][0]
image = cv2.imread(path)
# Create a HOG face detector using the built-in dlib class
# Load the image into an array
start=time.time()
try: faces_cnn=face_detector( image , 1 )
except: continue
count += 1
gray = cv2.cvtColor(image, cv2.COLOR_BGR2GRAY)
for face in faces_cnn :
fa = FaceAligner(predictor,desiredLeftEye=(0.25, 0.25), desiredFaceWidth=256) # 얼굴 Crop Size에서 눈 사이 간격과 width 비율 = 1: 1-0.25-0.25 = 1:0.5
faceAligned = fa.align(image,gray,face.rect)
cv2.imwrite('/home/daehyeon/hdd/processed/fake_256/{}.jpg'.format(count), faceAligned)
# cv2.imshow("Aligned", faceAligned)
end=time.time()
print('{}개 중 {}번째 이미지'.format(len(train_data.imgs),count), '걸린시간:' , format( end - start , '.2f' ) )
cv2.waitKey()
cv2.destroyAllWindows()
face_detector = dlib.cnn_face_detection_model_v1(
"mmod_human_face_detector.dat")
else:
# faster face detector
face_detector = dlib.get_frontal_face_detector()
#init the face landmarks detector
predictor_5_face_landmarks = dlib.shape_predictor(
"shape_predictor_5_face_landmarks.dat")
#init the object tracker
object_tracker = dlib.correlation_tracker()
# tight face aligner : padding = 0.2
aligner_tight = FaceAligner(
face_size=112,
face_padding=0.2,
predictor_5_face_landmarks=predictor_5_face_landmarks)
# loose face aligner : padding = 0.4
aligner_loose = FaceAligner(
face_size=112,
face_padding=0.4,
predictor_5_face_landmarks=predictor_5_face_landmarks)
# init the Face tracker
face_tracker = FaceTracker(face_detector, object_tracker)
aligners = []
for video_path in input_videos:
video_key = os.path.basename(video_path)
#set the out folder of the tight aligner