def __init__(self, dataset, *args, **kwargs):
kwargs = dict(kwargs, skip_images=True)
if dataset == '7Scenes':
from seven_scenes import SevenScenes
self.real_dset = SevenScenes(*args, real=True, **kwargs)
self.gt_dset = SevenScenes(*args, real=False, **kwargs)
elif dataset == 'RobotCar':
from robotcar import RobotCar
self.real_dset = RobotCar(*args, real=True, **kwargs)
self.gt_dset = RobotCar(*args, real=False, **kwargs)
else:
raise NotImplementedError
transforms.Resize(256),
transforms.RandomCrop(crop_size),
transforms.ToTensor()
])
# dataset loader
data_dir = osp.join('..', 'data', 'deepslam_data', args.dataset)
kwargs = dict(sequence=args.scene,
data_path=data_dir,
train=True,
real=False,
transform=data_transform)
if args.dataset == '7Scenes':
dset = SevenScenes(**kwargs)
elif args.dataset == 'RobotCar':
dset = RobotCar(**kwargs)
else:
raise NotImplementedError
# accumulate
batch_size = 8
num_workers = batch_size
loader = DataLoader(dset,
batch_size=batch_size,
num_workers=num_workers,
collate_fn=safe_collate)
acc = np.zeros((3, crop_size[0], crop_size[1]))
sq_acc = np.zeros((3, crop_size[0], crop_size[1]))
for batch_idx, (imgs, _) in enumerate(loader):
imgs = imgs.numpy()
acc += np.sum(imgs, axis=0)
steps=steps,
skip=skip,
real=real,
variable_skip=variable_skip,
include_vos=args.pose_graph,
vo_func=vo_func,
no_duplicates=False,
**kwargs)
L = len(data_set.dset)
elif args.dataset == '7Scenes':
from dataset_loaders.seven_scenes import SevenScenes
data_set = SevenScenes(**kwargs)
L = len(data_set)
elif args.dataset == 'RobotCar':
from dataset_loaders.robotcar import RobotCar
data_set = RobotCar(**kwargs)
L = len(data_set)
else:
raise NotImplementedError
# loader (batch_size MUST be 1)
batch_size = 1
assert batch_size == 1
loader = DataLoader(data_set,
batch_size=batch_size,
shuffle=False,
num_workers=5,
pin_memory=True)
# activate GPUs
CUDA = torch.cuda.is_available()
# datasets
data_dir = osp.join('..', 'data', 'deepslam_data', args.dataset)
kwargs = dict(scene=args.scene,
data_path=data_dir,
transform=data_transform,
target_transform=target_transform,
seed=seed)
if args.model == 'posenet':
if args.dataset == '7Scenes':
from dataset_loaders.seven_scenes import SevenScenes
train_set = SevenScenes(train=True, **kwargs)
val_set = SevenScenes(train=False, **kwargs)
elif args.dataset == 'RobotCar':
from dataset_loaders.robotcar import RobotCar
train_set = RobotCar(train=True, **kwargs)
val_set = RobotCar(train=False, **kwargs)
else:
raise NotImplementedError
elif args.model.find('mapnet') >= 0:
kwargs = dict(kwargs,
dataset=args.dataset,
skip=skip,
steps=steps,
variable_skip=variable_skip)
if args.model.find('++') >= 0:
train_set = MFOnline(vo_lib=vo_lib,
gps_mode=(vo_lib == 'gps'),
**kwargs)
val_set = None
else:
'images')
parser.add_argument('--scene', type=str, required=True)
parser.add_argument('--n_cores', type=int, default=4)
parser.add_argument('--val', action='store_true')
args = parser.parse_args()
if args.val:
print('processing VAL data using {:d} cores'.format(args.n_cores))
else:
print('processing TRAIN data using {:d} cores'.format(args.n_cores))
# create data loader
batch_size = args.n_cores
data_dir = osp.join('..', 'data', 'deepslam_data', 'RobotCar')
transform = transforms.Compose([transforms.Scale(256),
transforms.Lambda(lambda x: np.asarray(x))])
dset = RobotCar(scene=args.scene, data_path=data_dir, train=(not args.val),
transform=transform, undistort=True)
loader = DataLoader(dset, batch_size=batch_size, num_workers=args.n_cores,
collate_fn=safe_collate)
# gather information about output filenames
base_dir = osp.join(data_dir, args.scene)
if args.val:
split_filename = osp.join(base_dir, 'test_split.txt')
else:
split_filename = osp.join(base_dir, 'train_split.txt')
with open(split_filename, 'r') as f:
seqs = [l.rstrip() for l in f if not l.startswith('#')]
im_filenames = []
for seq in seqs:
seq_dir = osp.join(base_dir, seq)
def __init__(self,
dataset,
include_vos=False,
no_duplicates=False,
*args,
**kwargs):
"""
:param steps: Number of frames to return on every call
:param skip: Number of frames to skip
:param variable_skip: If True, skip = [1, ..., skip]
:param include_vos: True if the VOs have to be appended to poses. If real
and include_vos are both on, it gives absolute poses from GT and VOs from
the SLAM / DSO
:param no_duplicates: if True, does not duplicate frames when len(self) is
not a multiple of skip*steps
"""
self.steps = kwargs.pop('steps', 2)
self.skip = kwargs.pop('skip', 1)
self.variable_skip = kwargs.pop('variable_skip', False)
self.real = kwargs.pop('real', False)
self.include_vos = include_vos
self.train = kwargs['train']
self.vo_func = kwargs.pop('vo_func', calc_vos_simple)
self.no_duplicates = no_duplicates
if dataset == '7Scenes':
from dataset_loaders.seven_scenes import SevenScenes
self.dset = SevenScenes(*args, real=self.real, **kwargs)
if self.include_vos and self.real:
self.gt_dset = SevenScenes(*args,
skip_images=True,
real=False,
**kwargs)
elif dataset == 'InLoc':
from dataset_loaders.inloc import InLoc
self.dset = InLoc(*args, real=self.real, **kwargs)
if self.include_vos and self.real:
# This might not work
self.gt_dset = InLoc(*args,
skip_images=True,
real=False,
**kwargs)
elif dataset == 'InLocRes':
from dataset_loaders.inloc import InLocQuery
self.dset = InLocQuery(*args, real=self.real, **kwargs)
if self.include_vos and self.real:
# This might not work
self.gt_dset = InLoc(*args,
skip_images=True,
real=False,
**kwargs)
elif dataset == 'RobotCar':
from dataset_loaders.robotcar import RobotCar
self.dset = RobotCar(*args, real=self.real, **kwargs)
if self.include_vos and self.real:
self.gt_dset = RobotCar(*args,
skip_images=True,
real=False,
**kwargs)
else:
raise NotImplementedError
self.L = self.steps * self.skip