def __init__(self, root, list_path, set='val',
max_iters=None,
crop_size=(321, 321), mean=(128, 128, 128),
load_labels=True,
info_path=DEFAULT_INFO_PATH, labels_size=None,
label_folder='',
num_classes=19):
self.label_folder = label_folder
super().__init__(root, list_path, set, max_iters, crop_size, labels_size, mean)
self.load_labels = load_labels
if num_classes==16:
self.info = json_load(INFO_PATH_16)
elif num_classes==7:
self.info = json_load(INFO_PATH_7)
else:
self.info = json_load(info_path)
self.mapping = np.array(self.info['label2train'], dtype=np.int)
self.map_vector = np.zeros((self.mapping.shape[0],), dtype=np.int64)
for source_label, target_label in self.mapping:
self.map_vector[source_label] = target_label
def __init__(self,
root,
list_path,
set='val',
max_iters=None,
crop_size=(321, 321),
mean=(128, 128, 128),
load_labels=True,
info_path=DEFAULT_INFO_PATH,
labels_size=None):
super().__init__(root, list_path, set, max_iters, crop_size,
labels_size, mean)
self.load_labels = load_labels
self.info = json_load(info_path)
self.class_names = np.array(self.info['label'], dtype=np.str)
self.mapping = np.array(self.info['label2train'], dtype=np.int)
self.map_vector = np.zeros((self.mapping.shape[0], ), dtype=np.int64)
for source_label, target_label in self.mapping:
self.map_vector[source_label] = target_label
def __init__(self,
root,
list_path,
set='train',
max_iters=None,
crop_size=(321, 321),
mean=(128, 128, 128),
info_path=DEFAULT_INFO_PATH,
class_mappings=MAPILLARY_2_CITYSCAPES,
class_list=COMMON_7CLASSES,
scale_label=True):
super().__init__(root, list_path, set, max_iters, crop_size, None,
mean)
self.info = json_load(info_path)
self.map_vector = None
self.scale_label = scale_label
ori_class_names = [label['readable'] for label in self.info['labels']]
if class_mappings is not None:
self.map_vector = array_from_class_mappings(
ori_class_names, class_mappings, class_list)
self.class_names = list(class_list.keys())
def __init__(self, root, set='train', max_iters=None, crop_size=(321, 321), mean=(128, 128, 128),
class_mappings=classes_mappings_mapillary_to_cityscapes, model_classes=classes_ids,
load_instances=False, scale_label=True, labels_size=None, info_path=DEFAULT_INFO_PATH,
label_folder=''):
# root = Path(download('mapillary'))
self.label_folder = label_folder
self.path = Path(root) / set
self.crop_size = crop_size
if labels_size is None:
self.labels_size = crop_size
else:
self.labels_size = labels_size
self.load_instances = load_instances
self.mean = mean
self.scale_label = scale_label
self.info = json_load(info_path)
if self.label_folder is not '':
sorted_paths = (sorted((self.path / 'images').iterdir()),
sorted((Path(self.label_folder)).iterdir()),
sorted((self.path / 'instances').iterdir()))
else:
sorted_paths = map(lambda x: sorted((self.path / x).iterdir()),
('images', 'labels', 'instances'))
self.data_paths = list(zip(*sorted_paths))
if max_iters is not None:
self.data_paths = self.data_paths * int(np.ceil(float(max_iters) / len(self.data_paths)))
self.labels = json.loads((Path(root) / 'config.json').read_text())['labels']
self.vector_mappings = None
if class_mappings is not None:
dataset_classes = [label['readable'] for label in self.labels]
self.vector_mappings = array_from_class_mappings(dataset_classes,
class_mappings,
model_classes)
