def __init__(self, *args, **kwargs):
super(FacialFeatureTrainer, self).__init__(self, *args, **kwargs)
self.train_size = kwargs.pop('train_size', -1)
self.images = utils.load_samples(paths.TRAINING_IMAGE_PATH, ".png",
allow_consecutive=False)
self.landmarks = utils.load_samples(paths.TRAINING_LANDMARK_PATH,
".txt", allow_consecutive=False)
if self.train_size < 0:
self.train_size = len(self.images)
self.train()
def __init__(self, *args, **kwargs):
super(FacialFeatureTrainer, self).__init__(self, *args, **kwargs)
self.train_size = kwargs.pop('train_size', -1)
self.images = utils.load_samples(paths.TRAINING_IMAGE_PATH,
".png",
allow_consecutive=False)
self.landmarks = utils.load_samples(paths.TRAINING_LANDMARK_PATH,
".txt",
allow_consecutive=False)
if self.train_size < 0:
self.train_size = len(self.images)
self.train()
def generate_features_from_session(self, session, auid):
images_dir = os.path.join(paths.TRAINING_IMAGE_PATH, session)
landmarks_dir = os.path.join(paths.TRAINING_LANDMARK_PATH, session)
images = utils.load_samples(images_dir, ".png")
landmarks = utils.load_samples(landmarks_dir, ".txt")
if not images or not landmarks:
print "Invalid session %s on AU %d" % (session, auid)
exit()
initial_image = cv2.imread(images[0])
final_image = cv2.imread(images[-1])
gray, face, init_scale = utils.preprocess_face_image(initial_image)
if not face:
if auid == 1 or not MULTITHREAD:
self.say("\t%d faces found. Discarding." % (init_scale,))
return None
gray, face, final_scale = utils.preprocess_face_image(final_image)
if not face:
if auid == 1 or not MULTITHREAD:
self.say("\t%d faces found. Discarding." % (final_scale,))
return None
all_initial_landmarks = utils.load_landmarks(landmarks[0], init_scale)
all_final_landmarks = utils.load_landmarks(landmarks[-1], final_scale)
initial_landmarks = []
final_landmarks = []
for i in range(len(all_initial_landmarks)):
if i in EXAMINED_POINTS:
initial_landmarks.append(all_initial_landmarks[i])
final_landmarks.append(all_final_landmarks[i])
distances = utils.distances(initial_landmarks, final_landmarks)
return distances
def generate_features_from_session(self, session, auid):
images_dir = os.path.join(paths.TRAINING_IMAGE_PATH, session)
landmarks_dir = os.path.join(paths.TRAINING_LANDMARK_PATH, session)
images = utils.load_samples(images_dir, ".png")
landmarks = utils.load_samples(landmarks_dir, ".txt")
if not images or not landmarks:
print "Invalid session %s on AU %d" % (session, auid)
exit()
initial_image = cv2.imread(images[0])
final_image = cv2.imread(images[-1])
gray, face, init_scale = utils.preprocess_face_image(initial_image)
if not face:
if auid == 1 or not MULTITHREAD:
self.say("\t%d faces found. Discarding." % (init_scale, ))
return None
gray, face, final_scale = utils.preprocess_face_image(final_image)
if not face:
if auid == 1 or not MULTITHREAD:
self.say("\t%d faces found. Discarding." % (final_scale, ))
return None
all_initial_landmarks = utils.load_landmarks(landmarks[0], init_scale)
all_final_landmarks = utils.load_landmarks(landmarks[-1], final_scale)
initial_landmarks = []
final_landmarks = []
for i in range(len(all_initial_landmarks)):
if i in EXAMINED_POINTS:
initial_landmarks.append(all_initial_landmarks[i])
final_landmarks.append(all_final_landmarks[i])
distances = utils.distances(initial_landmarks, final_landmarks)
return distances
def set_sequence(self, sequence, type=".png"):
"""
Accepts a list of numpy arrays (good for webcam data) or a directory
name containing the images in the sequence (good for testing with
Cohn-Kanade)
"""
self.sequence = []
if not sequence:
return
if isinstance(sequence, basestring):
files = utils.load_samples(sequence, type)
images = []
for image in files:
image = cv2.imread(image)
images.append(image)
self.sequence = images
elif isinstance(sequence[0], np.array):
self.sequence = sequence