# datasets
data_dir = osp.join('..', 'data', 'deepslam_data', args.dataset)
kwargs = dict(dataset=args.dataset,
scene=args.scene,
data_path=data_dir,
transform=data_transform,
target_transform=target_transform,
reduce=args.reduce,
seed=seed)
if args.model == 'posenet':
train_set = Env(train=True, **kwargs)
val_set = Env(train=False, **kwargs)
elif args.model.find('mapnet') >= 0:
kwargs = dict(kwargs, skip=skip, steps=steps)
train_set = MF(train=True, **kwargs)
val_set = MF(train=False, **kwargs)
else:
raise NotImplementedError
# trainer
config_name = args.config_file.split('/')[-1]
config_name = config_name.split('.')[0]
if args.reduce is None:
experiment_name = '{:s}_{:s}_{:s}_{:s}'.format(args.dataset, args.scene,
args.model, config_name)
else:
experiment_name = '{:s}_{:s}_{:s}_{:s}_reduce'.format(
args.dataset, args.scene, args.model, config_name)
trainer = Trainer(model,
optimizer,
kwargs = dict(scene=args.scene,
data_path=data_dir,
train=train,
transform=data_transform,
target_transform=target_transform,
seed=seed)
if (args.model.find('mapnet') >= 0) or args.pose_graph:
if args.pose_graph:
assert real
kwargs = dict(kwargs, vo_lib=vo_lib)
vo_func = calc_vos_safe_fc if fc_vos else calc_vos_safe
data_set = MF(dataset=args.dataset,
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
input_types=input_types,
output_types=output_types,
train_split=train_split,
#concatenate_inputs=True
)
if args.dataset == 'AachenDayNight':
kwargs['night_augmentation'] = None
kwargs['resize'] = resize
if (args.model.find('mapnet') >= 0) or (args.model.find('semantic') >= 0) or (
args.model.find('multitask') >= 0):
vo_func = calc_vos_safe_fc if fc_vos else calc_vos_safe
data_set = MF(dataset=args.dataset,
steps=steps,
skip=skip,
real=real,
variable_skip=variable_skip,
include_vos=False,
vo_func=vo_func,
no_duplicates=False,
**kwargs)
L = len(data_set.dset)
if args.dataset == '7Scenes':
from dataset_loaders.seven_scenes import SevenScenes
data_set = SevenScenes(**kwargs)
elif args.dataset == 'DeepLoc':
from dataset_loaders.deeploc import DeepLoc
data_set = DeepLoc(**kwargs)
elif args.dataset == 'RobotCar':
from dataset_loaders.robotcar import RobotCar
data_set = RobotCar(**kwargs)
elif args.dataset == 'AachenDayNight':
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:
train_set = MF(train=True, real=real, **kwargs)
val_set = MF(train=False, real=real, **kwargs)
else:
raise NotImplementedError
# trainer
config_name = args.config_file.split('/')[-1]
config_name = config_name.split('.')[0]
experiment_name = '{:s}_{:s}_{:s}_{:s}'.format(args.dataset, args.scene,
args.model, config_name)
if args.learn_beta:
experiment_name = '{:s}_learn_beta'.format(experiment_name)
if args.learn_gamma:
experiment_name = '{:s}_learn_gamma'.format(experiment_name)
experiment_name += args.suffix
trainer = Trainer(model,
# read mean and stdev for un-normalizing predictions
pose_stats_file = osp.join(data_dir, args.scene, 'pose_stats.txt')
pose_m, pose_s = np.loadtxt(pose_stats_file) # max value
# dataset
train = not args.val
if train:
print 'Running {:s} on Training Dataset'.format(args.model)
else:
print 'Running {:s} on Validation Dataset'.format(args.model)
data_dir = osp.join('..', 'data', 'deepslam_data', args.dataset)
kwargs = dict(dataset=args.dataset, scene=args.scene, data_path=data_dir, train=train, transform=data_transform, target_transform=target_transform, reduce=args.reduce, seed=seed)
if args.model.find('mapnet') >= 0:
data_set = MF(steps=steps, skip=skip, **kwargs)
L = len(data_set.dset)
else:
data_set = Env(**kwargs)
L = len(data_set)
# 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=10, pin_memory=True)
# activate GPUs
CUDA = torch.cuda.is_available()
torch.manual_seed(seed)
if CUDA:
torch.cuda.manual_seed(seed)