def load_model(data_bunch, model_type, model_name):
"""
Function to create and load pretrained weights of convolutional learner
"""
learn = create_cnn(data_bunch, model_type, pretrained=False)
learn.load(model_name)
return learn
def __init__(self, data, backbone=None, pretrained_path=None):
super().__init__()
self._device = torch.device(
'cuda') if torch.cuda.is_available() else torch.device('cpu')
if not HAS_FASTAI:
_raise_fastai_import_error()
if backbone is None:
self._backbone = models.resnet34
elif type(backbone) is str:
self._backbone = getattr(models, backbone)
else:
self._backbone = backbone
self._emd_template = _EMD_TEMPLATE
self._code = feature_classifier_prf
self._data = data
self.learn = create_cnn(data, self._backbone, metrics=accuracy)
self.learn.model = self.learn.model.to(self._device)
if pretrained_path is not None:
self.load(pretrained_path)
def __init__(self,
data,
grids=[4, 2, 1],
zooms=[0.7, 1., 1.3],
ratios=[[1., 1.], [1., 0.5], [0.5, 1.]],
backbone=None,
drop=0.3,
bias=-4.,
focal_loss=False,
pretrained_path=None):
super().__init__()
self._device = torch.device(
'cuda') if torch.cuda.is_available else torch.device('cpu')
if backbone is None:
backbone = resnet34
self._create_anchors(grids, zooms, ratios)
ssd_head = SSDHead(grids,
self._anchors_per_cell,
data.c,
drop=drop,
bias=bias)
self._data = data
self.learn = create_cnn(data=data,
base_arch=backbone,
custom_head=ssd_head)
self.learn.model = self.learn.model.to(self._device)
if pretrained_path is not None:
self.load(pretrained_path)
if focal_loss:
self._loss_f = FocalLoss(data.c)
else:
self._loss_f = BCE_Loss(data.c)
self.learn.loss_func = self._ssd_loss
def get_learner(classes):
# TODO: Can we make this faster/lighter?
data = ImageDataBunch.single_from_classes(".", classes, ds_tfms=get_transforms(), size=224).normalize(imagenet_stats)
learn = create_cnn(data, resnet34, pretrained=False)
learn.load('makemodel-392')
return learn
def __init__(self,
data,
grids=[4, 2, 1],
zooms=[0.7, 1., 1.3],
ratios=[[1., 1.], [1., 0.5], [0.5, 1.]],
backbone=None,
drop=0.3,
bias=-4.,
focal_loss=False,
pretrained_path=None,
location_loss_factor=None):
super().__init__()
self._device = torch.device(
'cuda') if torch.cuda.is_available() else torch.device('cpu')
# assert (location_loss_factor is not None) or ((location_loss_factor > 0) and (location_loss_factor < 1)),
if location_loss_factor is not None:
if not ((location_loss_factor > 0) and (location_loss_factor < 1)):
raise Exception(
'`location_loss_factor` should be greater than 0 and less than 1'
)
self.location_loss_factor = location_loss_factor
if not HAS_FASTAI:
_raise_fastai_import_error()
if backbone is None:
self._backbone = models.resnet34
elif type(backbone) is str:
self._backbone = getattr(models, backbone)
else:
self._backbone = backbone
self._create_anchors(grids, zooms, ratios)
feature_sizes = model_sizes(create_body(self._backbone),
size=(data.chip_size, data.chip_size))
num_features = feature_sizes[-1][-1]
num_channels = feature_sizes[-1][1]
ssd_head = SSDHead(grids,
self._anchors_per_cell,
data.c,
num_features=num_features,
drop=drop,
bias=bias,
num_channels=num_channels)
self._data = data
self.learn = create_cnn(data=data,
arch=self._backbone,
custom_head=ssd_head)
self.learn.model = self.learn.model.to(self._device)
if pretrained_path is not None:
self.load(pretrained_path)
if focal_loss:
self._loss_f = FocalLoss(data.c)
else:
self._loss_f = BCE_Loss(data.c)
self.learn.loss_func = self._ssd_loss
def detect_facial_attributes(input_path, output_path, save_video):
path = Path(input_path)
# Creating a databunch
imagenet_stats = ([0.485, 0.456, 0.406], [0.229, 0.224, 0.225])
data = (ImageItemList.from_csv(
path, csv_name="labels.csv").no_split().label_from_df(
label_delim=" ").transform(
None,
size=128).databunch(no_check=True).normalize(imagenet_stats))
# Loading our model
learn = create_cnn(data, models.resnet50, pretrained=False)
learn.load("ff_stage-2-rn50")
# Loading HAAR cascade
face_cascade = cv2.CascadeClassifier("haarcascade_frontalface_default.xml")
cap = cv2.VideoCapture(0)
if save_video:
out = cv2.VideoWriter(output_path + "output.avi", -1, 20.0, (640, 480))
while True:
# Capture frame-by-frame
_, frame = cap.read()
# Our operations on the frame come here
gray = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
# Find faces using Haar cascade
face_coord = face_cascade.detectMultiScale(gray,
1.1,
5,
minSize=(30, 30))
## Looping through each face
for coords in face_coord:
## Finding co-ordinates of face
X, Y, w, h = coords
## Finding frame size
H, W, _ = frame.shape
## Computing larger face co-ordinates
X_1, X_2 = (max(0, X - int(w * 0.35)), min(X + int(1.35 * w), W))
Y_1, Y_2 = (max(0, Y - int(0.35 * h)), min(Y + int(1.35 * h), H))
## Cropping face and changing BGR To RGB
img_cp = frame[Y_1:Y_2, X_1:X_2].copy()
img_cp1 = cv2.cvtColor(img_cp, cv2.COLOR_BGR2RGB)
## Prediction of facial featues
prediction = str(
learn.predict(Image(pil2tensor(
img_cp1, np.float32).div_(255)))[0]).split(";")
label = (" ".join(prediction) if "Male" in prediction else
"Female " + " ".join(prediction))
label = (" ".join(prediction) if "No_Beard" in prediction else
"Beard " + " ".join(prediction))
## Drawing facial boundaries
cv2.rectangle(
img=frame,
pt1=(X, Y),
pt2=(X + w, Y + h),
color=(128, 128, 0),
thickness=2,
)
## Drawing facial attributes identified
label_list = label.split(" ")
for idx in range(1, len(label_list) + 1):
cv2.putText(
frame,
label_list[idx - 1],
(X, Y - 14 * idx),
cv2.FONT_HERSHEY_SIMPLEX,
0.45,
(0, 128, 0),
2,
)
# Display the resulting frame
cv2.imshow("frame", frame)
## Save the resulting frame
if save_video:
out.write(frame)
## Escape keys
if cv2.waitKey(1) & 0xFF == ord("q"):
break
# When everything done, release the capture
cap.release()
if save_video:
out.release()
cv2.destroyAllWindows()