def test_recognise_single_face(self):
for img_fn in one_face:
bboxes = pre_calculated_faces[os.path.basename(img_fn)]
logging.debug(
"Testing face recognition on file with a single face %s" %
(img_fn, ))
boxes = [BoundingBox(**b) for b in bboxes]
header = FaceRecognitionHeader(faces=boxes)
with open(img_fn, 'rb') as infile:
chunk = infile.read()
request = FaceRecognitionRequest(
header=header, image_chunk=ImageRGB(content=chunk))
result = self.stub.RecogniseFace(request)
# print(np.array([f for f in result.identities[0].identity]))
self.assertEqual(len(result.identities), len(bboxes))
def test_find_single_face(self):
for img_fn in one_face:
log.debug("Testing face detect %s on file with a single face %s" %
(
self.algorithm,
img_fn,
))
with open(img_fn, 'rb') as infile:
data = infile.read()
request = ImageRGB(content=data)
result = self.stub.FindFace(request)
log.debug("%s - %s - %s" %
(self.algorithm, img_fn, str(result.face_bbox)))
self.assertEqual(len(result.face_bbox), 1)
render_face_detect_debug_image(self, img_fn, result)
def test_recognise_no_faces(self):
for img_fn in no_faces:
# When there is no face, then this checks things don't explode when we give it a face bbox with no face
bboxes = list(pre_calculated_faces.values())[0]
logging.debug("Testing face recognition on file with no faces %s" %
(img_fn, ))
boxes = [BoundingBox(**b) for b in bboxes]
header = FaceRecognitionHeader(faces=boxes)
with open(img_fn, 'rb') as infile:
chunk = infile.read()
request = FaceRecognitionRequest(
header=header, image_chunk=ImageRGB(content=chunk))
result = self.stub.RecogniseFace(request)
# Should still have the right number of responses, even if they are meaningless
# print(np.array([f for f in result.identities[0].identity]))
self.assertEqual(len(result.identities), len(bboxes))
def test_find_multiple_faces(self):
for img_fn in multiple_faces:
log.debug("Testing face detect %s on file with multiple faces %s" %
(
self.algorithm,
img_fn,
))
with open(img_fn, 'rb') as infile:
data = infile.read()
request = ImageRGB(content=data)
result = self.stub.FindFace(request)
if img_fn.endswith('classroom_in_tanzania.jpg'):
log.debug("Haar cascade is known to fail on %s" % (img_fn, ))
else:
self.assertGreater(len(result.face_bbox), 1)
log.debug(str(result.face_bbox))
render_face_detect_debug_image(self, img_fn, result)
def test_align_multiple_faces(self):
for img_fn in multiple_faces:
bboxes = pre_calculated_faces[os.path.basename(img_fn)]
logging.debug(
"Testing face alignment on file with multiple faces %s" %
(img_fn, ))
boxes = [BoundingBox(**b) for b in bboxes]
header = FaceAlignmentHeader(source_bboxes=boxes)
with open(img_fn, 'rb') as infile:
chunk = infile.read()
request = FaceAlignmentRequest(
header=header, image_chunk=ImageRGB(content=chunk))
result = self.stub.AlignFace(request)
images = []
for i in result.image_chunk:
images.append(bytes(i.content))
self.assertEqual(len(images), len(bboxes))
def read_in_chunks(filename, face_bboxes, model="68", chunk_size=1024*64):
bboxes = []
for bbox in face_bboxes:
bboxes.append(BoundingBox(**bbox))
header = FaceRecognitionHeader(faces=bboxes)
flm = FaceRecognitionRequest(header=header)
yield flm
with open(filename, 'rb') as infile:
while True:
chunk = infile.read(chunk_size)
if chunk:
yield FaceRecognitionRequest(image_chunk=ImageRGB(content=chunk))
else:
# The chunk was empty, which means we're at the end
# of the file
return
def test_align_no_faces(self):
for img_fn in no_faces:
# When there is no face, then this checks things don't explode when we give it a face bbox with no face
bboxes = list(pre_calculated_faces.values())[0]
logging.debug("Testing face alignment on file with no faces %s" %
(img_fn, ))
boxes = [BoundingBox(**b) for b in bboxes]
header = FaceAlignmentHeader(source_bboxes=boxes)
with open(img_fn, 'rb') as infile:
chunk = infile.read()
request = FaceAlignmentRequest(
header=header, image_chunk=ImageRGB(content=chunk))
result = self.stub.AlignFace(request)
images = []
for i in result.image_chunk:
images.append(bytes(i.content))
# Should still have the right number of responses, even if they are meaningless
self.assertEqual(len(images), len(bboxes))
def test_get_landmarks_no_faces(self):
for img_fn in no_faces:
# When there is no face, then this checks things don't explode when we give it a face bbox with no face
bboxes = list(pre_calculated_faces.values())[0]
logging.debug("Testing face detect on file with no faces %s" %
(img_fn, ))
boxes = [BoundingBox(**b) for b in bboxes]
header = FaceLandmarkHeader(
landmark_model=self.algorithm,
faces=FaceDetections(face_bbox=boxes))
with open(img_fn, 'rb') as infile:
chunk = infile.read()
request = FaceLandmarkRequest(
header=header, image_chunk=ImageRGB(content=chunk))
result = self.stub.GetLandmarks(request)
# Should still have the right number of responses, even if they are meaningless
self.assertEqual(len(result.landmarked_faces), len(bboxes))
for i in range(0, len(bboxes)):
self.assertEqual(len(result.landmarked_faces[i].point),
int(self.algorithm))
def read_in_chunks(filename, source_bboxes, chunk_size=1024 * 64):
boxes = [BoundingBox(**b) for b in source_bboxes]
#source_p2d = [Point2D(x=int(p[0]), y=int(p[1])) for p in source_pts]
#target_p2d = [Point2D(x=int(p[0]), y=int(p[1])) for p in target_pts]
#source = FaceLandmarks(point=source_p2d)
#target = FaceLandmarks(point=target_p2d)
header = services.grpc.face_alignment_pb2.FaceAlignmentHeader(
source_bboxes=boxes, )
yield services.grpc.face_alignment_pb2.FaceAlignmentRequest(header=header)
with open(filename, 'rb') as infile:
while True:
chunk = infile.read(chunk_size)
if chunk:
yield services.grpc.face_alignment_pb2.FaceAlignmentRequest(
image_chunk=ImageRGB(content=chunk))
else:
# The chunk was empty, which means we're at the end
# of the file
return
def test_get_landmarks_single_face(self):
for img_fn in one_face:
bboxes = pre_calculated_faces[os.path.basename(img_fn)]
logging.debug(
"Testing face landmark prediction %s with a single face %s"
% (
self.algorithm,
img_fn,
))
boxes = [BoundingBox(**b) for b in bboxes]
header = FaceLandmarkHeader(
landmark_model=self.algorithm,
faces=FaceDetections(face_bbox=boxes))
with open(img_fn, 'rb') as infile:
chunk = infile.read()
request = FaceLandmarkRequest(
header=header, image_chunk=ImageRGB(content=chunk))
result = self.stub.GetLandmarks(request)
self.assertEqual(len(result.landmarked_faces), len(bboxes))
self.assertEqual(len(result.landmarked_faces[0].point),
int(self.algorithm))
def AlignFace(self, request, context):
image_data = bytearray()
header = request.header
image_data.extend(bytes(request.image_chunk.content))
img_bytes = io.BytesIO(image_data)
img = ioimg.imread(img_bytes)
log.debug("Received image with shape %s" % str(img.shape))
# Drop alpha channel if it exists
if img.shape[-1] == 4:
img = img[:,:,:3]
log.debug("Dropping alpha channel from image")
#source_pts = np.float32([[p.x, p.y] for p in header.source.point])
#target_pts = np.float32([[p.x, p.y] for p in header.target.point])
aligned_faces = []
for bbox in header.source_bboxes:
raw_dst_img = do_alignment(img, bbox)
aligned_faces.append(ImageRGB(content=raw_dst_img))
return FaceAlignmentResponse(image_chunk=aligned_faces)
def AlignFace(self, request_iterator, context):
image_data = bytearray()
header = None
for i, data in enumerate(request_iterator):
if i == 0:
if data.HasField("header"):
header = data.header
continue
else:
raise Exception("No header provided!")
else:
image_data.extend(bytes(data.image_chunk.content))
img_bytes = io.BytesIO(image_data)
img = ioimg.imread(img_bytes)
log.debug("Received image with shape %s" % str(img.shape))
# Drop alpha channel if it exists
if img.shape[-1] == 4:
img = img[:, :, :3]
log.debug("Dropping alpha channel from image")
#source_pts = np.float32([[p.x, p.y] for p in header.source.point])
#target_pts = np.float32([[p.x, p.y] for p in header.target.point])
for bbox in header.source_bboxes:
raw_dst_img = do_alignment(img, bbox)
chunk_size = 1024 * 64
yield FaceAlignmentResponse(header=FaceAlignmentResponseHeader())
for i in range(0, len(raw_dst_img), chunk_size):
yield FaceAlignmentResponse(image_chunk=ImageRGB(
content=raw_dst_img[i:i + chunk_size]))
