def landmark(video, model, tracking, output):
"""Facial features detection"""
# face generator
frame_width, frame_height = video.frame_size
faceGenerator = getFaceGenerator(tracking,
frame_width, frame_height,
double=False)
faceGenerator.send(None)
face = Face(landmarks=model)
with open(output, 'w') as foutput:
for timestamp, rgb in video:
# get all detected faces at this time
T, faces = faceGenerator.send(timestamp)
# not that T might be differ slightly from t
# due to different steps in frame iteration
for identifier, boundingBox, _ in faces:
landmarks = face._get_landmarks(rgb, boundingBox)
foutput.write('{t:.3f} {identifier:d}'.format(
t=T, identifier=identifier))
for x, y in landmarks:
foutput.write(' {x:.5f} {y:.5f}'.format(x=x / frame_width,
y=y / frame_height))
foutput.write('\n')
foutput.flush()
def features(video, model, shape, output):
"""Openface FaceNet feature extraction"""
face = Face(size=96, openface=model)
# shape generator
frame_width, frame_height = video.frame_size
landmarkGenerator = getLandmarkGenerator(shape, frame_width, frame_height)
landmarkGenerator.send(None)
with open(output, 'w') as foutput:
for timestamp, rgb in video:
T, shapes = landmarkGenerator.send(timestamp)
for identifier, landmarks in shapes:
normalized_rgb = face._get_normalized(rgb, landmarks)
normalized_bgr = cv2.cvtColor(normalized_rgb,
cv2.COLOR_BGR2RGB)
openface = face._get_openface(normalized_bgr)
foutput.write('{t:.3f} {identifier:d}'.format(
t=T, identifier=identifier))
for x in openface:
foutput.write(' {x:.5f}'.format(x=x))
foutput.write('\n')
foutput.flush()
def extract(video, landmark_model, embedding_model, tracking, landmark_output,
embedding_output):
"""Facial features detection"""
# face generator
frame_width, frame_height = video.frame_size
faceGenerator = getFaceGenerator(tracking,
frame_width,
frame_height,
double=False)
faceGenerator.send(None)
face = Face(landmarks=landmark_model, embedding=embedding_model)
with open(landmark_output, 'w') as flandmark, \
open(embedding_output, 'w') as fembedding:
for timestamp, rgb in video:
# get all detected faces at this time
T, faces = faceGenerator.send(timestamp)
# not that T might be differ slightly from t
# due to different steps in frame iteration
for identifier, bounding_box, _ in faces:
landmarks = face.get_landmarks(rgb, bounding_box)
embedding = face.get_embedding(rgb, landmarks)
flandmark.write('{t:.3f} {identifier:d}'.format(
t=T, identifier=identifier))
for p in landmarks.parts():
x, y = p.x, p.y
flandmark.write(' {x:.5f} {y:.5f}'.format(
x=x / frame_width, y=y / frame_height))
flandmark.write('\n')
fembedding.write('{t:.3f} {identifier:d}'.format(
t=T, identifier=identifier))
for x in embedding:
fembedding.write(' {x:.5f}'.format(x=x))
fembedding.write('\n')
flandmark.flush()
fembedding.flush()
def extract(video, landmark_model, embedding_model, tracking, landmark_output, embedding_output):
"""Facial features detection"""
# face generator
frame_width, frame_height = video.frame_size
faceGenerator = getFaceGenerator(tracking,
frame_width, frame_height,
double=False)
faceGenerator.send(None)
face = Face(landmarks=landmark_model,
embedding=embedding_model)
with open(landmark_output, 'w') as flandmark, \
open(embedding_output, 'w') as fembedding:
for timestamp, rgb in video:
# get all detected faces at this time
T, faces = faceGenerator.send(timestamp)
# not that T might be differ slightly from t
# due to different steps in frame iteration
for identifier, bounding_box, _ in faces:
landmarks = face.get_landmarks(rgb, bounding_box)
embedding = face.get_embedding(rgb, landmarks)
flandmark.write('{t:.3f} {identifier:d}'.format(
t=T, identifier=identifier))
for p in landmarks.parts():
x, y = p.x, p.y
flandmark.write(' {x:.5f} {y:.5f}'.format(x=x / frame_width,
y=y / frame_height))
flandmark.write('\n')
fembedding.write('{t:.3f} {identifier:d}'.format(
t=T, identifier=identifier))
for x in embedding:
fembedding.write(' {x:.5f}'.format(x=x))
fembedding.write('\n')
flandmark.flush()
fembedding.flush()
