def __init__(self,
root_folder,
type,
input_type,
split_filename='',
transform_color=None,
transform_depth=None,
img_res=None,
crop_res=None,
for_autoencoding=False,
fpa_subj_split=False,
fpa_obj_split=False):
super(FPADatasetPoseRegressionFromVQVAE,
self).__init__(root_folder,
type,
input_type,
transform_color=transform_color,
transform_depth=transform_depth,
img_res=img_res,
split_filename=split_filename,
for_autoencoding=for_autoencoding)
self.fpa_subj_split = fpa_subj_split
self.fpa_obj_split = fpa_obj_split
if split_filename == '':
fpa_io.create_split_file(self.root_folder,
self.video_folder,
perc_train=0.7,
perc_valid=0.15,
only_with_obj_pose=False,
fpa_subj_split=fpa_subj_split,
fpa_obj_split=fpa_obj_split,
split_filename='fpa_split_subj.p')
self.split_filename = self.default_split_filename
self.dataset_split = fpa_io.load_split_file(self.root_folder,
self.split_filename)
def __init__(self,
root_folder,
type,
input_type,
transform_color=None,
transform_depth=None,
img_res=None,
crop_res=None,
split_filename='',
for_autoencoding=False):
super(FPADatasetTracking,
self).__init__(root_folder,
type,
input_type,
transform_color=transform_color,
transform_depth=transform_depth,
img_res=img_res,
crop_res=crop_res,
split_filename=split_filename,
for_autoencoding=for_autoencoding)
if not crop_res is None:
self.crop_res = crop_res
if self.split_filename == '':
fpa_io.create_split_file(self.root_folder,
self.video_folder,
perc_train=0.7,
perc_valid=0.15)
else:
self.dataset_split = fpa_io.load_split_file(
self.root_folder, self.split_filename)
def __init__(self,
root_folder,
type,
transform=None,
img_res=None,
split_filename=''):
self.root_folder = root_folder
self.transform = transform
self.img_res = img_res
self.split_filename = split_filename
self.type = type
if self.split_filename == '':
fpa_io.create_split_file(self.root_folder,
self.gt_folder,
num_train_seq=2,
actions=None)
else:
self.dataset_tuples = fpa_io.load_split_file(
self.root_folder, self.split_filename)
def __init__(self, params_dict):
super(FPADatasetObjRGBReconstruction, self).\
__init__(params_dict['root_folder'],
split_filename=params_dict['split_filename'],
img_res=params_dict['img_res'])
self.params_dict = params_dict
# create or load dataset split
if params_dict['split_filename'] == '':
fpa_io.create_split_file(
params_dict['dataset_root_folder'],
perc_train=0.8,
perc_valid=0.,
only_with_obj_pose=True,
fpa_subj_split=False,
fpa_obj_split=False,
split_filename=params_dict['split_filename'])
self.dataset_split = fpa_io.load_split_file(self.root_folder,
self.split_filename)
self.type = params_dict['type']
self.unnormalize = UnNormalize(self.transform.transforms[-1].mean,
self.transform.transforms[-1].std,
img=True)
parser.add_argument('--fpa-obj-split',
default=False,
action='store_true',
help='Whether to use the FPA paper cross-object split')
parser.add_argument('--all',
default=False,
action='store_true',
help='Create all splits')
args = parser.parse_args()
if args.all:
fpa_io.create_split_file(args.dataset_root_folder,
args.video_folder,
perc_train=0.7,
perc_valid=0.15,
only_with_obj_pose=False,
fpa_subj_split=True,
fpa_obj_split=False)
fpa_io.create_split_file(args.dataset_root_folder,
args.video_folder,
perc_train=0.7,
perc_valid=0.15,
only_with_obj_pose=False,
fpa_subj_split=False,
fpa_obj_split=True)
fpa_io.create_split_file(args.dataset_root_folder,
args.video_folder,
perc_train=0.7,
perc_valid=0.15,
only_with_obj_pose=False,
myprint("\tSaving a checkpoint...", log_filepath)
torch.save(state, filename)
loss_function = nn.NLLLoss()
actions = [
'charge_cell_phone', 'clean_glasses', 'close_juice_bottle',
'close_liquid_soap', 'close_milk', 'close_peanut_butter', 'drink_mug',
'flip_pages', 'flip_sponge', 'give_card'
]
fpa_io.create_split_file(args.dataset_root_folder,
args.gt_folder,
num_train_seq=2,
actions=[
'charge_cell_phone', 'clean_glasses',
'close_juice_bottle', 'close_liquid_soap',
'close_milk'
])
dataset_tuples = fpa_io.load_split_file(args.dataset_root_folder)
lstm_baseline = LSTMBaseline(num_joints=21,
num_actions=dataset_tuples['num_actions'],
use_cuda=args.use_cuda)
if args.use_cuda:
lstm_baseline = lstm_baseline.cuda()
optimizer = optim.Adadelta(lstm_baseline.parameters(),
rho=0.9,