def test_iou(self):
r1 = face_extraction.Rectangle(20, 20, centerX=30, centerY=30)
r2 = face_extraction.Rectangle(20, 20, centerX=40, centerY=30)
self.assertNotEqual(r1.IOU(r2), 1)
self.assertNotEqual(r1.IOU(r2), 0)
self.assertEqual(r1.IOU(r1), 1)
self.assertEqual(r2.IOU(r2), 1)
r1 = face_extraction.Rectangle(20, 20, centerX=10, centerY=10)
r2 = face_extraction.Rectangle(20, 20, centerX=20, centerY=10)
self.assertEqual(r1.IOU(r2), 1 / 3)
r2 = face_extraction.Rectangle(20, 20, centerX=50, centerY=20)
self.assertEqual(r1.IOU(r2), 0)
def decode_object(o):
if '__Point__' in o:
a = face_extraction.Point(0, 0)
a.__dict__.update(o['__Point__'])
return a
elif '__Rectangle__' in o:
a = face_extraction.Rectangle(10, 10, centerX=5, centerY=5)
a.__dict__.update(o['__Rectangle__'])
return a
elif '__FaceRect__' in o:
a = face_extraction.FaceRect(None, None, None, None)
a.__dict__.update(o['__FaceRect__'])
if a.encoding is not None:
a.encoding = np.asarray(a.encoding)
# if a.image is not None:
# a.image = np.asarray(a.image)
# else:
# logger.critical("Returned face did not have a regular image.")
# if a.square_face is not None:
# a.square_face = np.asarray(a.square_face)
# else:
# logger.critical("Returned face did not have a regular image.")
return a
elif '__datetime__' in o:
return datetime.strptime(o['__datetime__'], '%Y-%m-%dT%H:%M:%S')
return o
def test_resizing(self):
a = face_extraction.Rectangle(50, 43, leftEdge=10, topEdge=15)
# print(a)
a_size = 50 * 43
a.resize(2)
self.assertEqual(4 * a_size, a.height * a.width)
a.resize(0.5)
self.assertEqual(a_size, a.height * a.width)
with self.assertRaises(ValueError):
a.resize(0)
def test_missing_args(self):
with self.assertRaises(face_extraction.rectangleError):
a = face_extraction.Rectangle(50, 40)
with self.assertRaises(face_extraction.rectangleError):
a = face_extraction.Rectangle(50, 40, centerX=5)
with self.assertRaises(face_extraction.rectangleError):
a = face_extraction.Rectangle(50,
40,
centerX=5,
centerY=5,
leftEdge=3)
with self.assertRaises(face_extraction.rectangleError):
a = face_extraction.Rectangle(50,
40,
centerX=5,
centerY=5,
leftEdge=3,
topEdge=6)
with self.assertRaises(face_extraction.rectangleError):
a = face_extraction.Rectangle(50, 40, leftEdge=3)
def test_intersect(self):
r1 = face_extraction.Rectangle(20, 20, centerX=30, centerY=30)
r2 = face_extraction.Rectangle(20, 20, centerX=35, centerY=35)
ovd = r1.intersect(r2)
self.assertEqual(ovd, 225)
def test_make_rect(self):
a = face_extraction.Rectangle(50, 40, centerX=5, centerY=5)
b = face_extraction.Rectangle(50, 40, leftEdge=3, topEdge=6)
def associate_detections_and_tags(image,
pristine_image,
detected_faces,
tagged_faces,
disp_photo=False,
test=False):
# Top-level function that takes a list of face_recognition
# image pyramid detections (from my detect_pyramid function)
# as well as tags from the XMP metadata (i.e. Picasa tags)
# and associates them based on geometry.
# Make sure our detected faces are the appropriate type
if len(detected_faces) > 0:
for df in detected_faces:
assert isinstance(df, face_extraction.FaceRect)
# Make sure our tagged faces are appropriate types and
# have the right keys.
if len(tagged_faces) > 0:
for tf in tagged_faces:
assert isinstance(tf, dict)
assert 'Name' in tf.keys()
assert 'bounding_rectangle' in tf.keys()
# # Load in the pixels of the image, then copy it to
# # pristine_image (that won't get drawn on)
# image = face_recognition.load_image_file(
# image_path)
# pristine_image = copy.deepface_recognition.load_image_file(
# image_path)
# A test case to see if we can reject super-huge
# faces.
if test:
huge_face = {
'Name':
"nada",
'bounding_rectangle':
face_extraction.Rectangle(int(image.shape[0] * .95),
int(image.shape[1] * .95),
centerX=image.shape[1] // 2,
centerY=image.shape[0] // 2)
}
tagged_faces.append(huge_face)
if disp_photo:
draw_image = image.copy()
for df in detected_faces:
df.rectangle.drawOnPhoto(draw_image)
for tf in tagged_faces:
tf['bounding_rectangle'].drawOnPhoto(draw_image,
colorTriple=(0, 255, 0))
# Two steps -- we first need to merge the
# detected faces. Since we have an image pyramid
# approach to image detection, we want to first
# de-duplicate these before we try and mess with
# matching with Picasa images.
merged_detections = _merge_detected_faces(detected_faces, pristine_image)
if disp_photo:
for d in merged_detections:
d.rectangle.drawOnPhoto(draw_image, colorTriple=(0, 0, 255))
# After de-duplicating face_recognition tags, then
# we can merge them with Picasa tags.
fully_matched = _merge_detections_with_tags(merged_detections,
tagged_faces, pristine_image)
if disp_photo:
# A routine to write the names with the
# faces as well as make the tagged face boxes
# larger for drawing.
for df in fully_matched:
rr = df.rectangle.copy()
# Expand the rectangle +- 10 pixels in each direction
rr.left = rr.left - 10
rr.top = rr.top - 10
rr.width = rr.width + 20
rr.height = rr.height + 20
df.rectangle.drawOnPhoto(draw_image, colorTriple=(235, 189, 52))
font = cv2.FONT_HERSHEY_SIMPLEX
bottomLeftCornerOfText = (rr.left, rr.bottom)
fontScale = 2
fontColor = (255, 255, 255)
lineType = 4
if df.name is not None:
name = df.name
else:
name = "None"
cv2.putText(draw_image, name, bottomLeftCornerOfText, font,
fontScale, fontColor, lineType)
plt.figure()
plt.imshow(draw_image)
plt.show()
# Cut out the fully-matched faces from the pristine
# image and assign it to the image field of the
# FaceRect structure.
for df in fully_matched:
r = df.rectangle
cut_rectangle = pristine_image[r.top:r.bottom, r.left:r.right]
assert (df.face_image_nonrect.shape == cut_rectangle.shape)
df.face_image_nonrect = cut_rectangle
return fully_matched