def __init__(self):
"""
Class constructor.
"""
self._faceDet = FaceDetector()
'''
The instance of the face detector.
'''
self._bank = GaborBank()
'''
The instance of the bank of Gabor filters.
'''
self._emotionsDet = EmotionsDetector()
'''
The instance of the emotions detector.
'''
self._face = FaceData()
'''
Data of the last face detected.
'''
self._emotions = OrderedDict()
'''
def __dataloader(self, train):
# init data generators
transform = transforms.Compose([
transforms.ToTensor(),
# transforms.Normalize((0.1307,), (0.3081,))
])
# dataset = CustomDataset(transform=transform)
dataset = FaceData(root='dataset', train=train, transform=transform)
if not train:
device = torch.device(
"cuda:0" if torch.cuda.is_available() else "cpu")
self.support_set = torch.from_numpy(
dataset.support_set).float().to(device)
# when using multi-node (ddp) we need to add the datasampler
train_sampler = None
batch_size = self.hparams.batch_size
if self.use_ddp:
train_sampler = DistributedSampler(dataset)
should_shuffle = train_sampler is None
loader = DataLoader(dataset=dataset,
batch_size=batch_size,
shuffle=should_shuffle,
sampler=train_sampler,
num_workers=0,
drop_last=True)
return loader
def detect(self, image, downSampleRatio=None):
if FaceDetector._detector is None or FaceDetector._predictor is None:
FaceDetector._detector = dlib.get_frontal_face_detector()
faceModel = os.path.abspath('D:\\workspace\\emotional-analysis\\models\\face_model.dat' \
.format(os.path.dirname(__file__)))
FaceDetector._predictor = dlib.shape_predictor(faceModel)
if downSampleRatio is not None:
detImage = cv2.resize(image, (0, 0),
fx=1.0 / downSampleRatio,
fy=1.0 / downSampleRatio)
else:
detImage = image
# 이미지에서 얼굴을 감지
detectedFaces = FaceDetector._detector(detImage, 1)
if len(detectedFaces) == 0:
return False, None
# 많은 얼굴이 발견 되더라도 첫 번째 얼굴만 고려
region = detectedFaces[0]
if downSampleRatio is not None:
region = dlib.rectangle(region.left() * downSampleRatio,
region.top() * downSampleRatio,
region.right() * downSampleRatio,
region.bottom() * downSampleRatio)
# 얼굴 영역에 모양 모델을 맞추어 위치를 예측
faceShape = FaceDetector._predictor(image, region)
face = FaceData()
face.landmarks = np.array([[p.x, p.y] for p in faceShape.parts()])
margin = 10
x, y, w, h = cv2.boundingRect(face.landmarks)
face.region = (max(x - margin,
0), max(y - margin,
0), min(x + w + margin, image.shape[1] - 1),
min(y + h + margin, image.shape[0] - 1))
return True, face
def __init__(self):
# 얼굴 검출기 인스턴스
self._faceDet = FaceDetector()
# Gabor 인스턴스
self._bank = GaborBank()
# 감정 탐지기
self._emotionsDet = EmotionsDetector()
# 검출 된 마지막 얼굴 데이터
self._face = FaceData()
# 감지 된 마지막 검정 데이터
self._emotions = OrderedDict()
def detect(self, image, downSampleRatio = None):
"""
Tries to automatically detect a face in the given image.
This method uses the face detector/predictor from the dlib package (with
its default face model) to detect a face region and 68 facial landmarks.
Even though dlib is able to detect more than one face in the image, for
the current purposes of the fsdk project only a single face is needed.
Hence, only the biggest face detected (estimated from the region size)
is considered.
Parameters
------
image: numpy.array
Image data where to search for the face.
downSampleRatio: float
Returns
------
result: bool
Indication on the success or failure of the facial detection.
face: FaceData
Instance of the FaceData class with the region and landmarks of the
detected face, or None if no face was detected.
"""
#####################
# Setup the detector
#####################
# Initialize the static detector and predictor if this is first use
if FaceDetector._detector is None or FaceDetector._predictor is None:
FaceDetector._detector = dlib.get_frontal_face_detector()
faceModel = os.path.abspath('{}/models/face_model.dat' \
.format(os.path.dirname(__file__)))
FaceDetector._predictor = dlib.shape_predictor(faceModel)
#####################
# Performance cues
#####################
# If requested, scale down the original image in order to improve
# performance in the initial face detection
if downSampleRatio is not None:
detImage = cv2.resize(image, (0, 0), fx=1.0 / downSampleRatio,
fy=1.0 / downSampleRatio)
else:
detImage = image
#####################
# Face detection
#####################
# Detect faces in the image
detectedFaces = FaceDetector._detector(detImage, 1)
if len(detectedFaces) == 0:
return False, None
# No matter how many faces have been found, consider only the first one
region = detectedFaces[0]
# If downscaling was requested, scale back the detected region so the
# landmarks can be proper located on the image in full resolution
if downSampleRatio is not None:
region = dlib.rectangle(region.left() * downSampleRatio,
region.top() * downSampleRatio,
region.right() * downSampleRatio,
region.bottom() * downSampleRatio)
# Fit the shape model over the face region to predict the positions of
# its facial landmarks
faceShape = FaceDetector._predictor(image, region)
#####################
# Return data
#####################
face = FaceData()
# Update the object data with the predicted landmark positions and
# their bounding box (with a small margin of 10 pixels)
face.landmarks = np.array([[p.x, p.y] for p in faceShape.parts()])
margin = 10
x, y, w, h = cv2.boundingRect(face.landmarks)
face.region = (
max(x - margin, 0),
max(y - margin, 0),
min(x + w + margin, image.shape[1] - 1),
min(y + h + margin, image.shape[0] - 1)
)
return True, face