def read_paths_and_labels(self, labels_file):
paths = []
labels = []
try:
with open(labels_file, 'r') as file:
for line in file:
line_split = line.split(',')
paths.append(
os.path.join(
self.dirdata,
tools.adapt_path_to_current_os(line_split[0])))
labels.append(int(line_split[1]))
except FileNotFoundError as ex:
logging.error('File ' + labels_file + ' does not exist.')
logging.error(str(ex))
raise
# Remove data or shuffle:
if self.percent_of_data != 100:
# Remove data:
indexes = np.random.choice(np.arange(len(labels)),
int(self.percent_of_data / 100.0 *
len(labels)),
replace=False)
else:
# Shuffle data at least:
indexes = np.arange(len(labels))
if self.shuffle_data:
np.random.shuffle(indexes)
aux_paths = paths
aux_labels = labels
paths = []
labels = []
for i in range(len(indexes)):
paths.append(aux_paths[indexes[i]])
labels.append(aux_labels[indexes[i]])
# Remove the remaining examples that do not fit in a batch.
if len(paths) % self.batch_size != 0:
aux_paths = paths
aux_labels = labels
paths = []
labels = []
for i in range(
len(aux_paths) - (len(aux_paths) % self.batch_size)):
paths.append(aux_paths[i])
labels.append(aux_labels[i])
assert len(
paths
) % self.batch_size == 0, 'Number of images is not a multiple of batch size'
return paths, labels
def get_filenames(self, split):
if split != 'train' and split != 'val':
raise Exception('Split name not recognized.')
list_file = os.path.join(self.dirdata, split + '_files.txt')
try:
with open(list_file, 'r') as fid:
filenamesnoext = fid.read().splitlines()
for i in range(len(filenamesnoext)):
filenamesnoext[i] = tools.adapt_path_to_current_os(
filenamesnoext[i])
except FileNotFoundError as ex:
logging.error('File ' + list_file + ' does not exist.')
logging.error(str(ex))
raise
# Remove data or shuffle:
if self.opts.percent_of_data != 100:
# Remove data:
indexes = np.random.choice(np.arange(len(filenamesnoext)),
int(self.opts.percent_of_data / 100.0 *
len(filenamesnoext)),
replace=False)
else:
# Shuffle data at least:
indexes = np.arange(len(filenamesnoext))
if self.opts.shuffle_data:
np.random.shuffle(indexes)
aux = filenamesnoext
filenamesnoext = []
for i in range(len(indexes)):
filenamesnoext.append(aux[indexes[i]])
# Remove the remaining examples that do not fit in a batch.
if len(filenamesnoext) % self.batch_size != 0:
aux = filenamesnoext
filenamesnoext = []
for i in range(len(aux) - (len(aux) % self.batch_size)):
filenamesnoext.append(aux[i])
assert len(
filenamesnoext
) % self.batch_size == 0, 'Number of images is not a multiple of batch size'
return filenamesnoext
def read_paths_and_labels(labels_file, dirdata):
paths = []
labels = []
try:
with open(labels_file, 'r') as file:
for line in file:
line_split = line.split(',')
paths.append(
os.path.join(dirdata,
tools.adapt_path_to_current_os(
line_split[0])))
labels.append(int(line_split[1]))
except FileNotFoundError as ex:
print('File ' + labels_file + ' does not exist.')
print(str(ex))
raise
# Shuffle data:
indexes = np.arange(len(labels))
np.random.shuffle(indexes)
aux_paths = paths
aux_labels = labels
paths = []
labels = []
for i in range(len(indexes)):
paths.append(aux_paths[indexes[i]])
labels.append(aux_labels[indexes[i]])
# Remove the remaining examples that do not fit in a batch.
if len(paths) % batch_size != 0:
aux_paths = paths
aux_labels = labels
paths = []
labels = []
for i in range(len(aux_paths) - (len(aux_paths) % batch_size)):
paths.append(aux_paths[i])
labels.append(aux_labels[i])
assert len(
paths
) % batch_size == 0, 'Number of images is not a multiple of batch size'
return paths, labels