def smart_detect(self):
if self.context['detectors'] and self.context['smart']:
storage = self.context['storage']
engine = self.context['engine']
storage_key = '%s_%d_%d' % (self.context['image_url'], engine.size[0], engine.size[1])
if self.context['crop_left']:
storage_key = storage_key + '_%d_%d_%d_%d' % (self.context['crop_left'],
self.context['crop_top'],
self.context['crop_right'],
self.context['crop_bottom']
)
focal_points = storage.get_detector_data(storage_key)
if focal_points:
for point in focal_points:
self.context['focal_points'].append(FocalPoint.from_dict(point))
else:
detectors = self.context['detectors']
detectors[0](index=0, detectors=detectors).detect(self.context)
points = []
focal_points = self.context['focal_points']
for point in focal_points:
points.append(point.to_dict())
storage.put_detector_data(storage_key, points)
def after_smart_detect(self, focal_points=[], points_from_storage=False):
self.manual_crop()
self.calculate_target_dimensions()
for point in focal_points:
self.context.request.focal_points.append(
FocalPoint.from_dict(point))
if self.context.request.focal_points and self.context.modules.storage and not points_from_storage:
storage = self.context.modules.storage
points = []
for point in self.context.request.focal_points:
points.append(point.to_dict())
storage.put_detector_data(self.smart_storage_key, points)
self.adjust_focal_points()
if self.context.request.debug:
self.debug()
else:
if self.context.request.fit_in:
self.fit_in_resize()
else:
self.auto_crop()
self.resize()
self.flip()
self.done_callback()
def after_smart_detect(self, focal_points=[], points_from_storage=False):
self.manual_crop()
self.calculate_target_dimensions()
for point in focal_points:
self.context.request.focal_points.append(FocalPoint.from_dict(point))
if self.context.request.focal_points and self.context.modules.storage and not points_from_storage:
storage = self.context.modules.storage
points = []
for point in self.context.request.focal_points:
points.append(point.to_dict())
storage.put_detector_data(self.smart_storage_key, points)
self.adjust_focal_points()
if self.context.request.debug:
self.debug()
else:
if self.context.request.fit_in:
self.fit_in_resize()
else:
self.auto_crop()
self.resize()
self.flip()
self.done_callback()
async def do_smart_detection(self):
focal_points = await (
self.context.modules.storage.get_detector_data(self.smart_storage_key)
)
points_from_storage = focal_points is not None
if focal_points is None:
detectors = self.context.modules.detectors
focal_points = await detectors[0](
self.context, index=0, detectors=detectors
).detect()
if focal_points is not None:
for point in focal_points:
self.context.request.focal_points.append(FocalPoint.from_dict(point))
if (
self.context.request.focal_points
and self.context.modules.storage
and not points_from_storage
):
storage = self.context.modules.storage
points = []
for point in self.context.request.focal_points:
points.append(point.to_dict())
await storage.put_detector_data(self.smart_storage_key, points)
def detect(self, callback):
engine = self.context.modules.engine
try:
engine.image_data_as_rgb()
img = np.array(engine.image)
self.net.setInput(
cv2.dnn.blobFromImage(img, size=(300, 300), swapRB=True))
detections = self.net.forward()
except Exception as e:
logger.exception(e)
logger.warn(
'Error during feature detection; skipping to next detector')
self.next(callback)
return
confidence_threshold = 0.2
num_detections = 0
for detection in detections[0, 0, :, :]:
confidence = float(detection[2])
if confidence < confidence_threshold:
continue
num_detections += 1
class_id = int(detection[1]) - 1 # make it zero-indexed
class_name = coco_classes[class_id]
left = int(detection[3] * img.shape[1])
top = int(detection[4] * img.shape[0])
right = int(detection[5] * img.shape[1])
bottom = int(detection[6] * img.shape[0])
width = right - left
height = bottom - top
# If the detection is of a person,
# and the person is vertically oriented,
# this uses the upper 1/4 of the box to focus on the face.
# In the case the person is horizontal, perhaps reclining,
# then the focal point will remain at their center.
# In the case the person is upside down, this would focus on the feet instead of the face.
# But consider - whoever is publishing a picture of an upside down person
# might appreciate that it focuses on the feet.
if class_name == 'person' and height > width:
height = int(height * 0.25)
self.context.request.focal_points.append(
FocalPoint.from_dict({
'x':
left + (width / 2),
'y':
top + (height / 2),
'width':
width,
'height':
height,
'z':
confidence,
'origin':
'DNN Object Detection (class: {})'.format(class_name)
}))
if num_detections > 0:
callback()
else:
self.next(callback)
def test_new_point_to_dict(self):
point = FocalPoint.from_dict({'x': 10.1, 'y': 20.1, 'z': 5.1})
expect(point.to_dict()).to_be_like({
'x': 10.1,
'y': 20.1,
'z': 5.1,
'origin': 'alignment',
'width': 1.0,
'height': 1.0
})
def test_new_point_to_dict(self):
point = FocalPoint.from_dict({
"x": 10.1,
"y": 20.1,
"z": 5.1,
"width": 1.1,
"height": 1.6
})
expect(point.to_dict()).to_be_like({
"x": 10,
"y": 20,
"z": 5,
"origin": "alignment",
"width": 1,
"height": 1,
})
def after_smart_detect(self, focal_points=[], points_from_storage=False):
for point in focal_points:
self.context.request.focal_points.append(FocalPoint.from_dict(point))
if self.context.request.focal_points and self.context.modules.storage and not points_from_storage:
storage = self.context.modules.storage
points = []
for point in self.context.request.focal_points:
points.append(point.to_dict())
storage.put_detector_data(self.smart_storage_key, points)
if self.running_smart_detection:
self.should_run_image_operations = True
return
self.do_image_operations()
def detect(self, callback):
engine = self.context.modules.engine
try:
engine.image_data_as_rgb()
img = np.array(engine.image)
self.net.setInput(
cv2.dnn.blobFromImage(img,
size=(300, 300),
mean=(104., 177., 123.)))
faces = self.net.forward()
except Exception as e:
logger.exception(e)
logger.warn(
'Error during feature detection; skipping to next detector')
self.next(callback)
return
# TODO: choose threshold based on empirical evidence
confidence_threshold = 0.3
num_faces = 0
for face in faces[0, 0, :, :]:
confidence = float(face[2])
if confidence < confidence_threshold:
continue
num_faces += 1
left = int(face[3] * img.shape[1])
top = int(face[4] * img.shape[0])
right = int(face[5] * img.shape[1])
bottom = int(face[6] * img.shape[0])
width = right - left
height = bottom - top
self.context.request.focal_points.append(
FocalPoint.from_dict({
'x': left + (width / 2),
'y': top + (height / 2),
'width': width,
'height': height,
'z': confidence,
'origin': 'DNN Face Detection'
}))
if num_faces > 0:
callback()
else:
self.next(callback)
def test_new_point_to_dict(self):
point = FocalPoint.from_dict({'x': 10.1, 'y': 20.1, 'z': 5.1})
expect(point.to_dict()).to_be_like({'x': 10.1, 'y': 20.1, 'z': 5.1, 'origin': 'alignment', 'width': 1.0, 'height': 1.0})
def test_new_point_from_dict(self):
point = FocalPoint.from_dict({'x': 10.1, 'y': 20.1, 'z': 5.1})
expect(point.x).to_equal(10.1)
expect(point.y).to_equal(20.1)
expect(point.weight).to_equal(5.1)
def test_new_point_from_dict(self):
point = FocalPoint.from_dict({'x': 10.1, 'y': 20.1, 'z': 5.1})
expect(point.x).to_equal(10.1)
expect(point.y).to_equal(20.1)
expect(point.weight).to_equal(5.1)
def test_new_point_from_dict(self):
point = FocalPoint.from_dict({"x": 10, "y": 20, "z": 5})
expect(point.x).to_equal(10)
expect(point.y).to_equal(20)
expect(point.weight).to_equal(5)
def topic(self):
return FocalPoint.from_dict({'x': 10.1, 'y': 20.1, 'z': 5.1})
def callback(buffer):
if buffer is None:
self._error(404)
return
context = dict(
loader=self.loader,
engine=self.engine,
storage=self.storage,
buffer=buffer,
should_crop=should_crop,
crop_left=crop_left,
crop_top=crop_top,
crop_right=crop_right,
crop_bottom=crop_bottom,
fit_in=fit_in,
should_flip_horizontal=horizontal_flip,
width=width,
should_flip_vertical=vertical_flip,
height=height,
halign=halign,
valign=valign,
extension=extension,
focal_points=[]
)
self.engine.load(buffer, extension)
if meta:
context['engine'] = JSONEngine(self.engine, image)
if self.detectors and should_be_smart:
focal_points = self.storage.get_detector_data(image)
if focal_points:
for point in focal_points:
context['focal_points'].append(FocalPoint.from_dict(point))
else:
with tempfile.NamedTemporaryFile(suffix='.jpg') as temp_file:
jpg_buffer = buffer if extension in ('.jpg', '.jpeg') else self.engine.read('.jpg')
temp_file.write(jpg_buffer)
temp_file.seek(0)
context['file'] = temp_file.name
self.detectors[0](index=0, detectors=self.detectors).detect(context)
points = []
focal_points = context['focal_points']
for point in focal_points:
points.append(point.to_dict())
self.storage.put_detector_data(image, points)
Transformer(context).transform()
if meta:
content_type = 'text/javascript' if options.META_CALLBACK_NAME else 'application/json'
else:
content_type = CONTENT_TYPE[context['extension']]
self.set_header('Content-Type', content_type)
results = context['engine'].read(context['extension'])
self.write(results)
self.finish()
def topic(self):
return FocalPoint.from_dict({'x': 10.1, 'y': 20.1, 'z': 5.1})
