def run_test():
input_shape = (IMG_SIZE[0], IMG_SIZE[1], 1)
classifier = SevenEmotionsClassifier()
preprocessor = Preprocessor(classifier, input_shape=input_shape)
postProcessor = PostProcessor(classifier)
neuralNet = NeuralNet(input_shape, preprocessor=preprocessor, train=False)
face_detector = dlib.get_frontal_face_detector()
if TEST_TYPE == "image":
img = cv2.imread(TEST_IMAGE)
faces, rectangles = preprocessor.get_faces(img, face_detector)
predictions = []
for i in range(len(faces)):
face = preprocessor.sanitize(faces[i])
predictions.append(neuralNet.predict(face))
postProcessor = postProcessor(img, rectangles, predictions)
cv2.imshow("Image", img)
cv2.waitKey(0)
cv2.destroyAllWindows()
elif TEST_TYPE == "video":
process_video(TEST_VIDEO, preprocessor, postProcessor, neuralNet)
elif TEST_TYPE == "webcam":
process_video(-1, preprocessor, postProcessor, neuralNet)
def __init__(self,
classifier,
input_shape=None,
batch_size=32,
augmentation=False,
verbose=True):
Preprocessor.__init__(self, classifier, input_shape, batch_size,
augmentation, verbose)
self.predictor = dlib.shape_predictor(SHAPE_PREDICTOR_PATH)
self.feature_extractor = DlibFeatureExtractor(self.predictor)
def __init__(self,classifier, input_shape = None,batch_size=BATCH_SIZE,augmentation = False,verbose = True,max_sequence_length=6):
Preprocessor.__init__(self,classifier,input_shape,batch_size,augmentation,verbose)
self.max_sequence_length = max_sequence_length
self.datagenerator = ImageDataGenerator(
rotation_range = 40,
width_shift_range = 0.1,
height_shift_range = 0.1,
shear_range = 0.1,
zoom_range = 0.1,
horizontal_flip=True,
data_format="channels_last"
)
def run_train():
# input_shape = (IMG_SIZE[0],IMG_SIZE[1],1)
input_shape = INPUT_SHAPE
classifier = SevenEmotionsClassifier()
if (NETWORK_TYPE == "mi"):
preprocessor = MultiInputPreprocessor(classifier,
input_shape=input_shape,
augmentation=AUGMENTATION)
neuralNet = MultiInputNeuralNet(input_shape,
preprocessor=preprocessor,
train=True)
elif NETWORK_TYPE == "si":
preprocessor = Preprocessor(classifier,
input_shape=input_shape,
augmentation=AUGMENTATION)
neuralNet = NeuralNet(input_shape,
preprocessor=preprocessor,
train=True)
elif NETWORK_TYPE == "rnn":
preprocessor = SequencialPreprocessor(
classifier, input_shape=input_shape,
augmentation=AUGMENTATION)("dataset/ck-split")
neuralNet = LSTMNet(input_shape, preprocessor=preprocessor, train=True)
elif NETWORK_TYPE == "vgg":
preprocessor = Preprocessor(classifier,
input_shape=input_shape,
augmentation=AUGMENTATION)
neuralNet = VGGFaceEmopyNet(input_shape,
preprocessor=preprocessor,
train=True)
elif NETWORK_TYPE == "drnn":
preprocessor = DlibSequencialPreprocessor(
classifier, input_shape=input_shape,
augmentation=AUGMENTATION)("dataset/ck-split")
neuralNet = DlibLSTMNet(input_shape,
preprocessor=preprocessor,
train=True)
else:
process = EmopySequencialProcess(input_shape, 6)
process.process_video(
"/home/mtk/iCog/projects/emopy/test-videos/75Emotions.mp4")
neuralNet.train()
def __init__(self,
classifier,
input_shape=None,
batch_size=32,
augmentation=False,
verbose=True):
"""
Args:
classifier:
input_shape:
batch_size:
augmentation:
verbose:
"""
Preprocessor.__init__(self, classifier, input_shape, batch_size,
augmentation, verbose)
self.predictor = dlib.shape_predictor()
self.feature_extractor = DlibFeatureExtractor(self.predictor)
def __init__(self,
classifier,
input_shape=None,
batch_size=BATCH_SIZE,
augmentation=False,
verbose=True,
max_sequence_length=10):
"""
Args:
classifier:
input_shape:
batch_size:
augmentation:
verbose:
max_sequence_length:
"""
Preprocessor.__init__(self, classifier, input_shape, batch_size,
augmentation, verbose)
self.max_sequence_length = max_sequence_length
self.datagenerator = ImageDataGenerator(rotation_range=20,
zoom_range=0.2,
height_shift_range=0.2,
width_shift_range=0.2)
def __init__(self,input_shape):
self.classifier = SevenEmotionsClassifier()
self.input_shape = input_shape
self.preprocessor = Preprocessor(self.classifier,input_shape=self.input_shape,batch_size=BATCH_SIZE,augmentation=False,verbose=True)
self.neuralNet = NeuralNet(self.input_shape,self.preprocessor)
def get_network(args):
if args.emotions == "all":
classifier = SevenEmotionsClassifier()
elif args.emotions == "pos-neg":
classifier = PositiveNegetiveClassifier()
elif args.emotions == "pos-neu":
classifier = PositiveNeutralClassifier()
else:
raise Exception(
"emotions should be one of all,pos-neg or pos-neu. But it is " +
str(args.emotions))
input_shape = (48, 48, 1)
if args.net == "face":
preprocessor = Preprocessor(classifier,
input_shape=input_shape,
batch_size=args.batch,
augmentation=args.augmentation,
verbose=args.verbose)
lgr = EmopyLogger(["logs/log.txt"])
net = NeuralNet((48, 48, 1), preprocessor, logger=lgr, train=True)
return net
elif args.net == "dlib":
preprocessor = DlibInputPreprocessor(classifier,
input_shape=input_shape,
batch_size=args.batch,
augmentation=args.augmentation,
verbose=args.verbose)
lgr = EmopyLogger(["logs/log.txt"])
net = DlibPointsInputNeuralNet((48, 48, 1),
preprocessor,
logger=lgr,
train=True)
return net
elif args.net == "face+dlib":
preprocessor = MultiInputPreprocessor(classifier,
input_shape=input_shape,
batch_size=args.batch,
augmentation=args.augmentation,
verbose=args.verbose)
lgr = EmopyLogger(["logs/log.txt"])
net = MultiInputNeuralNet((48, 48, 1),
preprocessor,
logger=lgr,
train=True)
return net
elif args.net == "vgg-face":
input_shape = (48, 48, 3)
preprocessor = Preprocessor(classifier,
input_shape=input_shape,
batch_size=args.batch,
augmentation=args.augmentation,
verbose=args.verbose)
lgr = EmopyLogger(["logs/log.txt"])
net = VGGFaceEmopyNet((48, 48, 1),
preprocessor,
logger=lgr,
train=True)
return net
elif args.net == "rnn":
preprocessor = SequencialPreprocessor(
classifier,
input_shape=input_shape,
batch_size=args.batch,
augmentation=args.augmentation,
verbose=args.verbose,
max_sequence_length=args.sequence_length)
lgr = EmopyLogger(["logs/log.txt"])
net = LSTMNet(input_shape,
preprocessor=preprocessor,
logger=lgr,
train=True,
max_sequence_length=args.sequence_length)
return net
elif args.net == "dlib-rnn":
preprocessor = DlibSequencialPreprocessor(
classifier,
input_shape=input_shape,
batch_size=args.batch,
augmentation=args.augmentation,
verbose=args.verbose,
max_sequence_length=args.sequence_length)
lgr = EmopyLogger(["logs/log.txt"])
net = DlibLSTMNet(input_shape,
preprocessor=preprocessor,
logger=lgr,
train=True,
max_sequence_length=args.sequence_length)
return net
else:
raise Exception(
"net arg should be one of face,face+dlib,vgg-face,rnn or dlib-rnn, but it is "
+ str(args.net))