def run(args): global averages, running averages, running = True, [] start_time = time.clock() video = cv2.VideoCapture() q = [] detector = dlib.get_frontal_face_detector() front = frontalization.Front(args) classifiers = [] classifiers.append(Classifier(args, start_time, args.classes, args.resource_dir, (150,150,3), colour=True, local=False)) classifiers.append(Classifier(args, start_time, args.classes, args.resource_dir, (150, 150), colour=False, local=False)) classifiers.append(Classifier(args, start_time, args.classes, args.resource_dir, (150,150,3), colour=True, local=False)) classifiers.append(Classifier(args, start_time, args.classes, args.resource_dir, (150,150,3), colour=True, local=True)) classifiers.append(Classifier(args, start_time, args.classes, args.resource_dir, (150, 150), colour=False, local=False)) classifiers.append(Classifier(args, start_time, args.classes, args.resource_dir, (150, 150), colour=False, local=True)) if video.grab(): while running: _, frame = video.read() image = cv2.resize(frame, (150, 150, frame.shape[2])) detection = detector(image, 1) if detection: rgb_image = frame[detection.top():detection.bottom(), detection.left():detection.right()] lbp_image = feature.local_binary_pattern(cv2.cvtColor(rgb_image, cv2.COLOR_BGR2GRAY).astype(np.float64), 8, 1, 'uniform') frgb_image = front.frontalized(image) flbp_image = feature.local_binary_pattern(cv2.cvtColor(frgb_image, cv2.COLOR_BGR2GRAY).astype(np.float64), 8, 1, 'uniform') classifications = [] classifiers[0].classify(rgb_image, args.log + 'rgb') classifiers[1].classify(lbp_image, (args.log + 'lbp')) classifiers[2].classify(frgb_image, (args.log + 'frgb')) classifiers[3].classify(frgb_image, (args.log + 'lfrgb')) classifiers[4].classify(flbp_image, (args.log + 'flbp')) classifiers[5].classify(flbp_image, (args.log + 'lflbp')) result = np.zeros(args.classes) for classification in classifications: for i in range(len(classification)): result[i] += classification[i] / 6 temp = [] if len(q) < 10: q.insert(0, result) else: q.pop() q.insert(0, result) for i in range(len(q[0])): average = 0 for j in range(len(q)): average += q[j][i] / 10 temp.append(average) averages = temp main.log(args, averages) else: main.log(args, 'No Face Found', True) if sys.stdin in select.select([sys.stdin], [], [], 0)[0]: raw_input() break
def run(args):
detector = dlib.get_frontal_face_detector()
front = frontalization.Front(args)
global predictions, running, arguments
predictions, running = [], True
start_time = time.clock()
q = []
video = cv2.VideoCapture()
classifier = Classifier(args, start_time, 3, args.resource_dir,
[88, 88, 3], args.scope, '/' + args.model + '/')
classifier.load_model()
if video.grab():
while running:
_, image = video.read()
detection = detector(image, 1)
if detection:
rgb_image = image[detection.top():detection.bottom(),
detection.left():detection.right()]
rgb_image = cv2.resize(rgb_image, (88, 88))
front_image = front.frontalized(rgb_image)
prediction = classifier.classify(front_image)
predictions.append(prediction)
main.log(
args, '\n{:.5f}s'.format(time.clock() - start_time) +
' Prediction: ' + str(prediction))
main.log(
args, '\n{:.5f}s'.format(time.clock() - start_time) +
' Averages: ' + str(predictions))
def run(args):
detector = dlib.get_frontal_face_detector()
front = frontalization.Front(args)
global predictions, running, arguments
predictions, running = [], True
start_time = time.clock()
q = []
videoDevice = ALProxy('ALVideoDevice', args.ip, args.port)
captureDevice = videoDevice.subscribeCamera('Camera', 0, 2, 13, 10)
image = np.zeros((480, 640, 3), np.uint8)
classifier = Classifier(args, start_time, 3, args.resource_dir, [88, 88, 3], args.scope, '/'+args.model+'/')
classifier.load_model()
while running:
image = stream2cv(image, videoDevice.getImageRemote(captureDevice))
if not image:
detection = detector(image, 1)
if detection:
rgb_image = image[detection.top():detection.bottom(), detection.left():detection.right()]
rgb_image = cv2.resize(rgb_image, (88, 88))
front_image = front.frontalized(rgb_image)
prediction = classifier.classify(front_image)
predictions.append(prediction)
main.log(args, '\n{:.5f}s'.format(time.clock() - start_time) + ' Prediction: ' + str(prediction))
main.log(args, '\n{:.5f}s'.format(time.clock() - start_time) + ' Averages: ' + str(predictions))
def save_image(args, start_time, save, data, type):
detector, count = dlib.get_frontal_face_detector(), 0
front = frontalization.Front(args)
for image_file in data:
image = cv2.imread(image_file[0], cv2.IMREAD_COLOR)
detections = detector(image, 1)
for _, detection in enumerate(detections):
#try:
face = image[detection.top():detection.bottom(),
detection.left():detection.right()]
if face.shape[0] == 0:
continue
face = cv2.resize(face, (88, 88))
images = []
images.append(face)
images.append(hue(face, 5))
images.append(hue(face, -5))
images.append(noisy('sp', images[0]))
images.append(noisy('gauss', images[0]))
images.append(hue(noisy('sp', images[0]), 5))
images.append(hue(noisy('sp', images[0]), -5))
images.append(hue(noisy('gauss', images[0]), 5))
images.append(hue(noisy('gauss', images[0]), -5))
for _image in images:
cv2.imwrite(
save + '/rgb/' + str(image_file[1]) + '/' + str(count) +
'.jpg', _image)
#lbp_image = feature.local_binary_pattern(cv2.cvtColor(_image, cv2.COLOR_BGR2GRAY).astype(np.float64), 8, 1, 'uniform').astype(np.uint8)
#lbp_image *= (255 / lbp_image.max())
#cv2.imwrite(save + '/lbp/' + str(image_file[1]) + '/' + str(count + 1) + '.jpg', lbp_image)
frgb_image = front.frontalized(image)
if frgb_image:
cv2.imwrite(
save + '/frgb/' + str(image_file[1]) + '/' +
str(count + 2) + '.jpg',
cv2.resize(frgb_image, (88, 88)))
#flbp_image = feature.local_binary_pattern(cv2.cvtColor(frgb_image, cv2.COLOR_BGR2GRAY).astype(np.float64), 8, 1, 'uniform').astype(np.uint8)
#flbp_image *= (255 / flbp_image.max())
#cv2.imwrite(save + '/flbp/' + str(image_file[1]) + '/' + str(count + 3) + '.jpg', flbp_image)
count += 4
if count % 100 == 0:
main.log(
args, '{:.5f}'.format(time.clock() - start_time) +
's ' + str(count) + ' ' + type + ' images extracted')
#except:
#pass
main.log(args,
str(time.clock() - start_time) + ' ' + type + ' Images Extracted')
def classify(args):
start_time = time.clock()
detector = dlib.get_frontal_face_detector()
front = frontalization.Front(args)
image = cv2.resize(cv2.imread(args.image, cv2.IMREAD_COLOR), (150, 150))
detection = detector(image, 1)
for _, detection in enumerate(detection):
rgb_image = image[detection.top():detection.bottom(),
detection.left():detection.right()]
lbp_image = feature.local_binary_pattern(
cv2.cvtColor(rgb_image, cv2.COLOR_BGR2GRAY).astype(np.float64), 8,
1, 'uniform').astype(np.uint8)
frgb_image = front.frontalized(image)
flbp_image = feature.local_binary_pattern(
cv2.cvtColor(frgb_image, cv2.COLOR_BGR2GRAY).astype(np.float64), 8,
1, 'uniform')
main.log(
args,
str(time.clock() - start_time) +
' Image Representations Extracted')
classifications = []
rgb_classifier = Classifier(args,
start_time,
args.classes,
args.resource_dir,
rgb_image.shape,
'rgb',
local=False)
classifications.append(
rgb_classifier.classify(rgb_image, (args.model_name + '/rgb/')))
lbp_classifier = Classifier(args,
start_time,
args.classes,
args.resource_dir,
lbp_image.shape,
'lbp',
local=False)
classifications.append(
lbp_classifier.classify(lbp_image, (args.model_name + '/lbp/')))
frgb_classifier = Classifier(args,
start_time,
args.classes,
args.resource_dir,
frgb_image.shape,
'frgb',
local=False)
classifications.append(
frgb_classifier.classify(frgb_image, (args.model_name + '/frgb/')))
lfrgb_classifier = Classifier(args,
start_time,
args.classes,
args.resource_dir,
frgb_image.shape,
'lfrgb',
local=True)
classifications.append(
lfrgb_classifier.classify(frgb_image,
(args.model_name + '/lfrgb/')))
flbp_classifier = Classifier(args,
start_time,
args.classes,
args.resource_dir,
flbp_image.shape,
'flbp',
local=False)
classifications.append(
flbp_classifier.classify(flbp_image, (args.model_name + '/flbp/')))
lflbp_classifier = Classifier(args,
start_time,
args.classes,
args.resource_dir,
flbp_image.shape,
'lflbp',
local=True)
classifications.append(
lflbp_classifier.classify(flbp_image,
(args.model_name + '/lflbp/')))
result = np.zeros(args.classes)
for classification in classifications:
for i in range(len(classification)):
result[i] += classification[i] / 6
main.log(
args, '\n' + str(time.clock() - start_time) + ' ' + args.image +
' classified as ' + str(result), True)
return True
main.log(
args, '\n' + str(time.clock() - start_time) +
' Could not detect face in inputted image', True)