def KITTI_noc(root, transform=None, target_transform=None,
co_transform=None, split=80):
train_list, test_list = make_dataset(root,False,split)
train_dataset = ListDataset(root, train_list, transform, target_transform, co_transform, loader=KITTI_loader)
test_dataset = ListDataset(root, test_list, transform, target_transform, flow_transforms.CenterCrop((320,1216)), loader=KITTI_loader)
return train_dataset, test_dataset
def mpi_sintel_final(root, transform=None, target_transform=None,
co_transform=None, split=80):
train_list, test_list = make_dataset(root, split, 'final')
train_dataset = ListDataset(root, train_list, transform, target_transform, co_transform)
test_dataset = ListDataset(root, test_list, transform, target_transform, flow_transforms.CenterCrop((384,1024)))
return train_dataset, test_dataset
def KITTI_noc(root, transform=None, target_transform=None,
co_transform=None, split=None):
train_list, test_list = make_dataset(root, split, False)
train_dataset = ListDataset(root, train_list, transform, target_transform, co_transform, loader=KITTI_loader)
# All test sample are cropped to lowest possible size of KITTI images
test_dataset = ListDataset(root, test_list, transform, target_transform, flow_transforms.CenterCrop((370,1224)), loader=KITTI_loader)
return train_dataset, test_dataset
def mpi_sintel_clean(root,
transform=None,
target_transform=None,
co_transform=None,
split=80):
train_list, test_list = make_dataset(root, 'clean', split)
#print(train_list)
train_dataset = ListDataset(root, train_list, transform, target_transform,
co_transform)
test_dataset = ListDataset(root, test_list, transform, target_transform,
flow_transforms.CenterCrop((384, 1024)))
return train_dataset, test_dataset
def BSD500(root, transform=None, target_transform=None, val_transform=None,
co_transform=None, split=None):
train_list, val_list = make_dataset(root)
if val_transform ==None:
val_transform = transform
train_dataset = ListDataset(root, 'bsd500', train_list, transform,
target_transform, co_transform,
loader=BSD_loader, datatype = 'train')
val_dataset = ListDataset(root, 'bsd500', val_list, val_transform,
target_transform, flow_transforms.CenterCrop((320,320)),
loader=BSD_loader, datatype = 'val')
return train_dataset, val_dataset
def mpi_sintel_both(root,
transform=None,
target_transform=None,
co_transform=None,
split=80):
'''load images from both clean and final folders.
We cannot shuffle input, because it would very likely cause data snooping
for the clean and final frames are not that different'''
train_list1, test_list1 = make_dataset(root, 'clean', split)
train_list2, test_list2 = make_dataset(root, 'final', split)
train_dataset = ListDataset(root, train_list1 + train_list2, transform,
target_transform, co_transform)
test_dataset = ListDataset(root, test_list1 + test_list2, transform,
target_transform,
flow_transforms.CenterCrop((384, 1024)))
return train_dataset, test_dataset
def mpi_sintel_both(root,
transform=None,
target_transform=None,
co_transform=None,
split=None):
'''load images from both clean and final folders.
We cannot shuffle input, because it would very likely cause data snooping
for the clean and final frames are not that different'''
assert (
isinstance(split, str)
), 'To avoid data snooping, you must provide a static list of train/val when dealing with both clean and final.'
' Look at Sintel_train_val.txt for an example'
train_list1, test_list1 = make_dataset(root, split, 'clean')
train_list2, test_list2 = make_dataset(root, split, 'final')
train_dataset = ListDataset(root, train_list1 + train_list2, transform,
target_transform, co_transform)
test_dataset = ListDataset(root, test_list1 + test_list2, transform,
target_transform,
flow_transforms.CenterCrop((384, 1024)))
return train_dataset, test_dataset
