def __init__(self, show=False, model=None, folder=None, save_model=None):
self.show = show
self.frtool = FaceExtractor()
self.fr_recog = FaceIdentify(show)
if model is not None:
self.fr_recog.load(model)
elif folder is not None:
self.fr_recog.train(folder, save_model)
else:
print('ERROR')
exit()
class FaceIdentify:
def __init__(self, show):
self.frtool = FaceExtractor()
self.show = show
def load(self, model_file):
try:
feature = np.load(filename)
print('Model have {} samples'.format(feature.shape[0]))
self.model = feature
except IOError:
print('Wrong numpy format in file {}'.format(filename))
exit()
def identify(self, feature):
dist = np.linalg.norm(self.model - feature, axis=1)
return np.mean(dist)
def train(self, folder, save_model):
filenames = os.listdir(folder)
features = list()
project_root = os.getcwd()
print('Get {} files'.format(len(filenames)))
for filename in filenames:
filepath = os.path.join(project_root, folder, filename)
img = cv2.imread(filepath)
if img is not None:
feature = self.frtool.extract_feature(img, self.show)
if feature is not None:
features.append(feature)
else:
print('Cannot found face in {}'.format(filepath))
else:
print('Cannot read file {}'.format(filepath))
if len(features) == 0:
print('Cannot extract any feature from folder {}'.format(folder))
else:
features = np.array(features)
num_samples = features.shape[0]
print('Get {} samples'.format(num_samples))
if save_model is None:
np.save('model', features)
print('Save to model.npy')
else:
np.save(save_model, features)
print('save to {}'.format(save_model))
self.model = features
class Identify:
@classmethod
def load_from_model(cls, filename, show):
if not os.path.isfile(filename):
print('Wrong file path {}'.format(filename))
exit()
else:
return cls(model=filename)
@classmethod
def train_from_folder(cls, folder, save_model, show):
if not os.path.isdir(folder):
print('Wrong folder path {}'.format(folder))
exit()
else:
return cls(folder=folder, save_model=save_model, show=show)
def __init__(self, show=False, model=None, folder=None, save_model=None):
self.show = show
self.frtool = FaceExtractor()
self.fr_recog = FaceIdentify(show)
if model is not None:
self.fr_recog.load(model)
elif folder is not None:
self.fr_recog.train(folder, save_model)
else:
print('ERROR')
exit()
def predict_image(self, filename):
if len(filename) == 0:
return
elif not os.path.isfile(filename):
print('Cannot found test file {}'.format(filename))
return
img = cv2.imread(filename)
feature = self.frtool.extract_feature(img, self.show)
if feature is None:
print('Cannot found face in {}'.format(filename))
else:
dist = self.fr_recog.identify(feature)
print(dist)
def predict_from_stream(self, web_cam_idx):
if web_cam_idx == -1:
return
cam = cv2.VideoCapture(web_cam_idx)
if not cam.isOpened:
print('Cannot load cam stream')
else:
is_open = True
ret, frame = cam.read()
while ret and is_open:
ret, frame = cam.read()
feature = self.frtool.extract_feature(frame, show=False)
flip_frame = cv2.flip(frame, 1)
if feature is None:
is_open = show_image(flip_frame, text='No Face', is_block=False)
else:
avg_dist = self.fr_recog.identify(feature)
# print(dist)
text = 'Similarity: {:.5f}'.format(avg_dist)
is_open = show_image(flip_frame, text=text, is_block=False)
def __init__(self):
self.frtool = FaceExtractor()
class FaceClassifier:
def __init__(self):
self.frtool = FaceExtractor()
def recog(self, feature):
if len(feature.shape) == 1:
feature = feature.reshape(1, -1)
return self.model.predict(feature)
def recog_proba(self, feature):
if len(feature.shape) == 1:
feature = feature.reshape(1, -1)
proba = self.model.predict_proba(feature)
identified = self.model.classes_[np.argmax(proba)]
return identified, proba
def load(self, model_file):
try:
self.model = pickle.load(open(model_file, 'rb'))
except pickle.UnpicklingError as e:
print('Wrong pickle file {}'.format(model_file))
def train(self, root, save_model):
subfolders = {
folder: os.path.join(root, folder)
for folder in os.listdir(root)
}
users_encoding = dict()
for _id, folder in subfolders.items():
if not os.path.isdir(folder):
print('Not folder, skip {}'.format(folder))
else:
print('Doing {}'.format(folder))
files = [
os.path.join(folder, _file) for _file in os.listdir(folder)
]
user_encoding = list()
for _file in filter(os.path.isfile, files):
img = cv2.imread(_file)
if img is not None:
feature = self.frtool.extract_feature(img)
if feature is not None:
user_encoding.append(feature)
if len(user_encoding):
users_encoding[_id] = user_encoding
print('Finish {}'.format(folder))
X = np.concatenate(list(users_encoding.values()))
Y = np.array(
list(_id for _id, encodings in users_encoding.items()
for loop in range(len(encodings))))
clf = KNeighborsClassifier(algorithm='ball_tree')
clf.fit(X, Y)
self.model = clf
pickle.dump(clf, open(save_model, 'wb'))
def __init__(self, show):
self.frtool = FaceExtractor()
self.show = show
class Classifier:
@classmethod
def load_from_model(cls, filename, show):
if not os.path.isfile(filename):
print('Wrong model file path {}'.format(filename))
exit()
else:
return cls(model=filename, show=show)
@classmethod
def train_from_folder(cls, root, save_model, show):
return cls(folder=root, save_model=save_model, show=show)
def __init__(self, show=False, model=None, folder=None, save_model=None):
self.show = show
self.frtool = FaceExtractor()
self.fr_recog = FaceClassifier()
if model is not None:
self.fr_recog.load(model)
elif folder is not None:
self.fr_recog.train(folder, save_model)
else:
print('ERROR')
exit()
def predict_image(self, filename):
if len(filename) == 0:
return
elif not os.path.isfile(filename):
print('Cannot found test file {}'.format(filename))
return
img = cv2.imread(filename)
feature = self.frtool.extract_feature(img, self.show)
if feature is None:
print('Cannot found face in {}'.format(filename))
else:
result = self.fr_recog.recog_proba(feature)
# dist = self.fr_recog.identify(feature)
print(result)
# dist = np.linalg.norm(self.features-feature, axis=1)
# print(np.mean(dist))
def predict_from_stream(self, web_cam_idx):
if web_cam_idx == -1:
return
cam = cv2.VideoCapture(web_cam_idx)
if not cam.isOpened:
print('Cannot load cam stream')
else:
is_open = True
ret, frame = cam.read()
while ret and is_open:
ret, frame = cam.read()
feature = self.frtool.extract_feature(frame, show=False)
flip_frame = cv2.flip(frame, 1)
if feature is None:
is_open = show_image(flip_frame,
text='No Face',
is_block=False)
else:
user, proba = self.fr_recog.recog_proba(feature)
# print(user, proba)
# avg_dist = self.fr_recog.recog(feature)
# print(dist)
# text = 'Similarity: {:.5f}'.format(avg_dist)
text = 'User: {}-{}'.format(user, np.max(proba))
is_open = show_image(flip_frame, text=text, is_block=False)