def load_data(self):
'''
Loads data to use in training.
'''
self.face_instances = list()
self.non_face_instances = list()
# Load data
if self.dataset == FaceDetData.MIT_Face_Data:
self.face_instances.extend(
ImageIO.loadFaceDetectionImages(
self.data_directory + '/train/face', 1))
self.face_instances.extend(
ImageIO.loadFaceDetectionImages(
self.data_directory + '/test/face', 1))
self.non_face_instances.extend(
ImageIO.loadFaceDetectionImages(
self.data_directory + '/train/non-face', 0))
self.non_face_instances.extend(
ImageIO.loadFaceDetectionImages(
self.data_directory + '/test/non-face', 0))
else:
raise RuntimeError('FaceDetTest not assigned a valid dataset')
img_size = numpy.shape(self.face_instances[0][0])
self.face_detector = FaceDetector(self.iterations, img_size)
self.total_face_instances = len(self.face_instances)
self.total_non_face_instances = len(self.non_face_instances)
# Sample from each class to create train/dev/test partitions
self.face_trn_data = list()
self.non_face_trn_data = list()
self.face_dev_data = list()
self.non_face_dev_data = list()
self.face_tst_data = list()
self.non_face_tst_data = list()
# Calculate how many samples should be in each partition
num_face_trn = int((self.trn_part / 100.0) * len(self.face_instances))
num_non_face_trn = int(
(self.trn_part / 100.0) * len(self.non_face_instances))
num_face_dev = int((self.dev_part / 100.0) * len(self.face_instances))
num_non_face_dev = int(
(self.dev_part / 100.0) * len(self.non_face_instances))
num_face_tst = int((self.tst_part / 100.0) * len(self.face_instances))
num_non_face_tst = int(
(self.tst_part / 100.0) * len(self.non_face_instances))
# If any leftovers, give more to training partition
num_face_trn += len(self.face_instances) - sum(
(num_face_trn, num_face_dev, num_face_tst))
num_non_face_trn += len(self.non_face_instances) - sum(
(num_non_face_trn, num_non_face_dev, num_non_face_tst))
# Partition trn/dev/tst
def partition(src, dest, num):
i = 0
while i < num:
if len(src) > 0:
idx = random.randint(0, len(src) - 1)
dest.append(src.pop(idx))
i += 1
partition(self.face_instances, self.face_trn_data, num_face_trn)
partition(self.non_face_instances, self.non_face_trn_data,
num_non_face_trn)
partition(self.face_instances, self.face_dev_data, num_face_dev)
partition(self.non_face_instances, self.non_face_dev_data,
num_non_face_dev)
partition(self.face_instances, self.face_tst_data, num_face_tst)
partition(self.non_face_instances, self.non_face_tst_data,
num_non_face_tst)
def load_data(self):
'''
Loads data to use in training.
'''
self.face_instances = list()
self.non_face_instances = list()
# Load data
if self.dataset == FaceDetData.MIT_Face_Data:
self.face_instances.extend(ImageIO.loadFaceDetectionImages(self.data_directory + '/train/face', 1))
self.face_instances.extend(ImageIO.loadFaceDetectionImages(self.data_directory + '/test/face', 1))
self.non_face_instances.extend(ImageIO.loadFaceDetectionImages(self.data_directory + '/train/non-face', 0))
self.non_face_instances.extend(ImageIO.loadFaceDetectionImages(self.data_directory + '/test/non-face', 0))
else:
raise RuntimeError('FaceDetTest not assigned a valid dataset')
img_size = numpy.shape(self.face_instances[0][0])
self.face_detector = FaceDetector(self.iterations, img_size)
self.total_face_instances = len(self.face_instances)
self.total_non_face_instances = len(self.non_face_instances)
# Sample from each class to create train/dev/test partitions
self.face_trn_data = list()
self.non_face_trn_data = list()
self.face_dev_data = list()
self.non_face_dev_data = list()
self.face_tst_data = list()
self.non_face_tst_data = list()
# Calculate how many samples should be in each partition
num_face_trn = int( (self.trn_part/100.0) * len(self.face_instances) )
num_non_face_trn = int( (self.trn_part/100.0) * len(self.non_face_instances) )
num_face_dev = int( (self.dev_part/100.0) * len(self.face_instances) )
num_non_face_dev = int( (self.dev_part/100.0) * len(self.non_face_instances) )
num_face_tst = int( (self.tst_part/100.0) * len(self.face_instances) )
num_non_face_tst = int( (self.tst_part/100.0) * len(self.non_face_instances) )
# If any leftovers, give more to training partition
num_face_trn += len(self.face_instances) - sum((num_face_trn, num_face_dev, num_face_tst))
num_non_face_trn += len(self.non_face_instances) - sum((num_non_face_trn, num_non_face_dev, num_non_face_tst))
# Partition trn/dev/tst
def partition(src, dest, num):
i = 0
while i < num:
if len(src) > 0:
idx = random.randint(0, len(src)-1)
dest.append( src.pop(idx) )
i += 1
partition(self.face_instances, self.face_trn_data, num_face_trn)
partition(self.non_face_instances, self.non_face_trn_data, num_non_face_trn)
partition(self.face_instances, self.face_dev_data, num_face_dev)
partition(self.non_face_instances, self.non_face_dev_data, num_non_face_dev)
partition(self.face_instances, self.face_tst_data, num_face_tst)
partition(self.non_face_instances, self.non_face_tst_data, num_non_face_tst)
