anticipation/anticipation/configs/long_term_anti/epic/verb_anti_vpretrain/common.py [8:87]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
task = 'anticipation'
val_timestamps = [0.25, 0.5, 0.75]
train_timestamps = [0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8]

train_many_shot = False
gfb_loss_weight = 0.0

train_cfg = None
test_cfg = None

# dataset settings
dataset_type = None
mode = "" # _verb|_noun|''
split = "S1"
pretrain = "verb"
feature_file = "data/{}/features/".format(dset) + "{{}}_lfb_s30_{}.pkl".format(pretrain)
fb_file = "data/{}/features/".format(dset) + "{{}}_lfb_s30_{}.pkl".format(pretrain)
graph_root = 'data/{}/graphs/'.format(dset)
node_num_member = 8
rand_visit = False
graph_aug = True
graph_drop = 0.5
gfb_loss_weight = 0.0


data = dict(
    videos_per_gpu=256,
    workers_per_gpu=4,
    train=dict(
        type=dataset_type,
        ann_file='data/{}/split/train_{}{}.csv'.format(dset, split, mode),
        label=label,
        task=task,
        dset=dset,
        fb=fb_file.format("train"),
        graph_root=graph_root,
        train_timestamps=train_timestamps,
        val_timestamps=val_timestamps,
        train_many_shot=train_many_shot,
        node_num_member=node_num_member,
        rand_visit=rand_visit,
        graph_aug=graph_aug,
        graph_drop=graph_drop,
        test_mode=False),
    val=dict(
        type=dataset_type,
        ann_file='data/{}/split/val_{}{}.csv'.format(dset, split, mode),
        label=label,
        task=task,
        dset=dset,
        fb=fb_file.format("val"),
        graph_root=graph_root,
        train_timestamps=train_timestamps,
        val_timestamps=val_timestamps,
        train_many_shot=train_many_shot,
        node_num_member=node_num_member,
        test_mode=True),
    test=dict(
        type=dataset_type,
        ann_file='data/{}/split/val_{}{}.csv'.format(dset, split, mode),
        label=label,
        task=task,
        dset=dset,
        fb=fb_file.format("val"),
        train_timestamps=train_timestamps,
        val_timestamps=val_timestamps,
        train_many_shot=train_many_shot,
        graph_root=graph_root,
        node_num_member=node_num_member,
        test_mode=True)
    )


# optimizer
optimizer = dict(type='Adam', lr=1e-3, weight_decay=1e-5)
optimizer_config = dict()

# learning policy
lr_config = dict(
    policy='step',
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -



anticipation/anticipation/configs/long_term_anti/gtea/verb_anti_vpretrain/common.py [8:86]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
task = 'anticipation'
val_timestamps = [0.25, 0.5, 0.75]
train_timestamps = [0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8]

train_many_shot = False
gfb_loss_weight = 0.0

train_cfg = None
test_cfg = None

# dataset settings
dataset_type = None
mode = "" # _verb|_noun|''
split = "S1"
pretrain = "verb"
feature_file = "data/{}/features/".format(dset) + "{{}}_lfb_s30_{}.pkl".format(pretrain)
fb_file = "data/{}/features/".format(dset) + "{{}}_lfb_s30_{}.pkl".format(pretrain)
graph_root = 'data/{}/graphs/'.format(dset)
node_num_member = 8
rand_visit = False
graph_aug = True
graph_drop = 0.5
gfb_loss_weight = 0.0

data = dict(
    videos_per_gpu=256,
    workers_per_gpu=4,
    train=dict(
        type=dataset_type,
        ann_file='data/{}/split/train_{}{}.csv'.format(dset, split, mode),
        label=label,
        task=task,
        dset=dset,
        fb=fb_file.format("train"),
        graph_root=graph_root,
        train_timestamps=train_timestamps,
        val_timestamps=val_timestamps,
        train_many_shot=train_many_shot,
        node_num_member=node_num_member,
        rand_visit=rand_visit,
        graph_aug=graph_aug,
        graph_drop=graph_drop,
        test_mode=False),
    val=dict(
        type=dataset_type,
        ann_file='data/{}/split/val_{}{}.csv'.format(dset, split, mode),
        label=label,
        task=task,
        dset=dset,
        fb=fb_file.format("val"),
        graph_root=graph_root,
        train_timestamps=train_timestamps,
        val_timestamps=val_timestamps,
        train_many_shot=train_many_shot,
        node_num_member=node_num_member,
        test_mode=True),
    test=dict(
        type=dataset_type,
        ann_file='data/{}/split/val_{}{}.csv'.format(dset, split, mode),
        label=label,
        task=task,
        dset=dset,
        fb=fb_file.format("val"),
        train_timestamps=train_timestamps,
        val_timestamps=val_timestamps,
        train_many_shot=train_many_shot,
        graph_root=graph_root,
        node_num_member=node_num_member,
        test_mode=True)
    )


#optimizer
optimizer = dict(type='Adam', lr=1e-3, weight_decay=1e-5)
optimizer_config = dict()

# learning policy
lr_config = dict(
    policy='step',
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -