def __init__(self,
root=Path.db_root_dir('VOCSegmentation'),
split='val',
transform=None,
download=True,
ignore_classes=[]):
# Set paths
self.root = root
valid_splits = ['trainaug', 'train', 'val']
assert (split in valid_splits)
self.split = split
if split == 'trainaug':
_semseg_dir = os.path.join(self.root, 'SegmentationClassAug')
else:
_semseg_dir = os.path.join(self.root, 'SegmentationClass')
_image_dir = os.path.join(self.root, 'images')
# Download
if download:
self._download()
# Transform
self.transform = transform
# Splits are pre-cut
print("Initializing dataloader for PASCAL VOC12 {} set".format(''.join(
self.split)))
split_file = os.path.join(self.root, 'sets', self.split + '.txt')
self.images = []
self.semsegs = []
with open(split_file, "r") as f:
lines = f.read().splitlines()
for ii, line in enumerate(lines):
# Images
_image = os.path.join(_image_dir, line + ".jpg")
assert os.path.isfile(_image)
self.images.append(_image)
# Semantic Segmentation
_semseg = os.path.join(_semseg_dir, line + '.png')
assert os.path.isfile(_semseg)
self.semsegs.append(_semseg)
assert (len(self.images) == len(self.semsegs))
# Display stats
print('Number of dataset images: {:d}'.format(len(self.images)))
# List of classes which are remapped to ignore index.
# This option is used for comparing with other works that consider only a subset of the pascal classes.
self.ignore_classes = [
self.VOC_CATEGORY_NAMES.index(class_name)
for class_name in ignore_classes
]
def get_train_dataset(p, transform=None):
if p['train_db_name'] == 'VOCSegmentation':
from data.dataloaders.pascal_voc import VOCSegmentation
return VOCSegmentation(root=Path.db_root_dir(p['train_db_name']),
saliency=p['train_db_kwargs']['saliency'],
transform=transform)
else:
raise ValueError('Invalid train db name {}'.format(p['train_db_name']))
def _download(self):
_fpath = os.path.join(Path.db_root_dir(), self.FILE)
if os.path.isfile(_fpath):
print('Files already downloaded')
return
else:
print('Downloading dataset from google drive')
mkdir_if_missing(os.path.dirname(_fpath))
download_file_from_google_drive(self.GOOGLE_DRIVE_ID, _fpath)
# extract file
cwd = os.getcwd()
print('\nExtracting tar file')
tar = tarfile.open(_fpath)
os.chdir(Path.db_root_dir())
tar.extractall()
tar.close()
os.chdir(cwd)
print('Done!')
def __init__(self,
root=Path.db_root_dir('VOCSegmentation'),
split='val',
transform=None,
download=True):
# Set paths
self.root = root
valid_splits = ['trainaug', 'train', 'val']
assert (split in valid_splits)
self.split = split
if split == 'trainaug':
_semseg_dir = os.path.join(self.root, 'SegmentationClassAug')
else:
_semseg_dir = os.path.join(self.root, 'SegmentationClass')
_image_dir = os.path.join(self.root, 'images')
# Download
if download:
self._download()
# Transform
self.transform = transform
# Splits are pre-cut
print("Initializing dataloader for PASCAL VOC12 {} set".format(''.join(
self.split)))
split_file = os.path.join(self.root, 'sets', self.split + '.txt')
self.images = []
self.semsegs = []
with open(split_file, "r") as f:
lines = f.read().splitlines()
for ii, line in enumerate(lines):
# Images
_image = os.path.join(_image_dir, line + ".jpg")
assert os.path.isfile(_image)
self.images.append(_image)
# Semantic Segmentation
_semseg = os.path.join(_semseg_dir, line + '.png')
assert os.path.isfile(_semseg)
self.semsegs.append(_semseg)
assert (len(self.images) == len(self.semsegs))
# Display stats
print('Number of dataset images: {:d}'.format(len(self.images)))
def __init__(self,
root=Path.db_root_dir('VOCSegmentation'),
saliency='supervised_model',
download=True,
transform=None,
overfit=False):
super(VOCSegmentation, self).__init__()
self.root = root
self.transform = transform
if download:
self._download()
self.images_dir = os.path.join(self.root, 'images')
valid_saliency = ['supervised_model', 'unsupervised_model']
assert (saliency in valid_saliency)
self.saliency = saliency
self.sal_dir = os.path.join(self.root, 'saliency_' + self.saliency)
self.images = []
self.sal = []
with open(os.path.join(self.root, 'sets/trainaug.txt'), 'r') as f:
all_ = f.read().splitlines()
for f in all_:
_image = os.path.join(self.images_dir, f + ".jpg")
_sal = os.path.join(self.sal_dir, f + ".png")
if os.path.isfile(_image) and os.path.isfile(_sal):
self.images.append(_image)
self.sal.append(_sal)
assert (len(self.images) == len(self.sal))
if overfit:
n_of = 32
self.images = self.images[:n_of]
self.sal = self.sal[:n_of]
# Display stats
print('Number of images: {:d}'.format(len(self.images)))