def __init__(self, mode, quality='fine', joint_transform_list=None,
img_transform=None, label_transform=None, eval_folder=None):
super(Loader, self).__init__(quality=quality, mode=mode,
joint_transform_list=joint_transform_list,
img_transform=img_transform,
label_transform=label_transform)
######################################################################
# Cityscapes-specific stuff:
######################################################################
self.root = cfg.DATASET.CITYSCAPES_DARK_DIR
self.id_to_trainid = cityscapes_labels.label2trainid
self.trainid_to_name = cityscapes_labels.trainId2name
self.fill_colormap()
img_ext = 'jpg'
mask_ext = '.png'
img_root = path.join(self.root, 'leftImg8bit')
mask_root = path.join(self.root, 'gtFine')
if mode == 'folder':
self.all_imgs = make_dataset_folder(eval_folder)
else:
self.fine_cities = cities_cv_split(self.root, mode, cfg.DATASET.CV)
self.all_imgs = self.find_cityscapes_images(
self.fine_cities, img_root, mask_root, img_ext, mask_ext)
logx.msg(f'cn num_classes {self.num_classes}')
self.fine_centroids = uniform.build_centroids(self.all_imgs,
self.num_classes,
self.train,
cv=cfg.DATASET.CV,
id2trainid=self.id_to_trainid)
self.centroids = self.fine_centroids
if cfg.DATASET.COARSE_BOOST_CLASSES and mode == 'train':
self.coarse_cities = coarse_cities(self.root)
img_root = path.join(self.root,
'leftImg8bit_trainextra/leftImg8bit')
mask_root = path.join(self.root, 'gtCoarse', 'gtCoarse')
self.coarse_imgs = self.find_cityscapes_images(
self.coarse_cities, img_root, mask_root, img_ext, mask_ext,
fine_coarse='gtCoarse')
if cfg.DATASET.CLASS_UNIFORM_PCT:
custom_coarse = (cfg.DATASET.CUSTOM_COARSE_PROB is not None)
self.coarse_centroids = uniform.build_centroids(
self.coarse_imgs, self.num_classes, self.train,
coarse=(not custom_coarse), custom_coarse=custom_coarse,
id2trainid=self.id_to_trainid)
for cid in cfg.DATASET.COARSE_BOOST_CLASSES:
self.centroids[cid].extend(self.coarse_centroids[cid])
else:
self.all_imgs.extend(self.coarse_imgs)
self.build_epoch()
def __init__(self,
mode,
quality='semantic',
joint_transform_list=None,
img_transform=None,
label_transform=None,
eval_folder=None):
super(Loader, self).__init__(quality=quality,
mode=mode,
joint_transform_list=joint_transform_list,
img_transform=img_transform,
label_transform=label_transform)
root = cfg.DATASET.CITYSURFACES_DIR
self.id_to_trainid = label2trainid
self.trainid_to_name = trainId2name
self.fill_colormap()
######################################################################
# Assemble image lists
######################################################################
if mode == 'folder':
img_ext = 'png'
mask_ext = 'png'
img_root = os.path.join(root, 'val', 'images')
mask_root = os.path.join(root, 'val', 'annotations')
self.all_imgs = self.find_images(img_root, mask_root, img_ext,
mask_ext)
elif mode == 'test':
self.all_imgs = make_dataset_folder(eval_folder, test_mode=True)
else:
splits = {'train': 'train', 'val': 'val', 'test': 'tests'}
split_name = splits[mode]
img_ext = 'png'
mask_ext = 'png'
img_root = os.path.join(root, split_name, 'images')
mask_root = os.path.join(root, split_name, 'annotations')
self.all_imgs = self.find_images(img_root, mask_root, img_ext,
mask_ext)
logx.msg('all imgs {}'.format(len(self.all_imgs)))
self.fine_centroids = uniform.build_centroids(
self.all_imgs,
self.num_classes,
self.train,
cv=cfg.DATASET.CV,
id2trainid=self.id_to_trainid)
self.centroids = self.fine_centroids
self.build_epoch()
def __init__(self,
mode,
quality='semantic',
joint_transform_list=None,
img_transform=None,
label_transform=None,
eval_folder=None):
super(Loader, self).__init__(quality=quality,
mode=mode,
joint_transform_list=joint_transform_list,
img_transform=img_transform,
label_transform=label_transform)
root = cfg.DATASET.VAIHINGEN_DIR
self.mode = mode
self.fill_colormap()
######################################################################
# Assemble image lists
######################################################################
if mode == 'folder':
self.all_imgs = make_dataset_folder(eval_folder)
else:
splits = {
'train': 'training',
'val': 'validation',
'trainval': 'trainval',
'test': 'testing'
}
split_name = splits[mode]
#img_ext = 'jpg'
#mask_ext = 'png'
#img_root = os.path.join(root, split_name, 'images')
#mask_root = os.path.join(root, split_name, 'labels')
if mode is 'test':
self.all_imgs = make_dataset(root, quality, mode)
else:
self.all_imgs = make_dataset(root, quality, mode)
logx.msg('all imgs {}'.format(len(self.all_imgs)))
self.centroids = uniform.build_centroids(self.all_imgs,
self.num_classes,
self.train,
cv=cfg.DATASET.CV)
self.build_epoch()
def __init__(self,
mode,
quality='fine',
joint_transform_list=None,
img_transform=None,
label_transform=None,
eval_folder=None):
super(Loader, self).__init__(quality=quality,
mode=mode,
joint_transform_list=joint_transform_list,
img_transform=img_transform,
label_transform=label_transform)
root = cfg.DATASET.KITTI_DIR
######################################################################
# Assemble image lists
######################################################################
if mode == 'folder':
self.all_imgs = make_dataset_folder(eval_folder)
else:
splits = {
'train': 'training',
'val': 'training',
'test': 'training'
}
split_name = splits[mode]
img_ext = 'png'
mask_ext = 'png'
img_root = os.path.join(root, split_name, 'image_2')
mask_root = os.path.join(root, split_name, 'semantic_rgb')
self.all_imgs = self.find_images(img_root, mask_root, img_ext,
mask_ext)
import ipdb
ipdb.set_trace()
self.fill_colormap()
logx.msg('all imgs {}'.format(len(self.all_imgs)))
self.centroids = uniform.build_centroids(self.all_imgs,
self.num_classes,
self.train,
cv=cfg.DATASET.CV)
self.build_epoch()
def __init__(self,
mode,
quality='semantic',
joint_transform_list=None,
img_transform=None,
label_transform=None,
eval_folder=None):
super(Loader, self).__init__(quality=quality,
mode=mode,
joint_transform_list=joint_transform_list,
img_transform=img_transform,
label_transform=label_transform)
root = cfg.DATASET.MAPILLARY_DIR
config_fn = os.path.join(root, 'config.json')
self.fill_colormap_and_names(config_fn)
######################################################################
# Assemble image lists
######################################################################
if mode == 'folder':
self.all_imgs = make_dataset_folder(eval_folder)
else:
splits = {
'train': 'training',
'val': 'validation',
'test': 'testing'
}
split_name = splits[mode]
img_ext = 'jpg'
mask_ext = 'png'
img_root = os.path.join(root, split_name, 'images')
mask_root = os.path.join(root, split_name, 'labels')
self.all_imgs = self.find_images(img_root, mask_root, img_ext,
mask_ext)
logx.msg('all imgs {}'.format(len(self.all_imgs)))
self.centroids = uniform.build_centroids(self.all_imgs,
self.num_classes,
self.train,
cv=cfg.DATASET.CV)
self.build_epoch()