forked from sair-lab/AirLoop
-
Notifications
You must be signed in to change notification settings - Fork 1
/
main.py
86 lines (67 loc) · 3.4 KB
/
main.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
#!/usr/bin/env python3
from pathlib import Path
import configargparse
from main_single import run
def main(args, extra_args):
# default environment orders
if args.envs is None:
if args.dataset == "tartanair":
args.envs = ["carwelding", "neighborhood", "office2", "abandonedfactory_night", "westerndesert"]
args.epochs = [4, 1, 2, 2, 3] if args.epochs is None else args.epochs
elif args.dataset == "nordland":
args.envs = ["spring", "summer", "fall", "winter"]
args.epochs = [3] if args.epochs is None else args.epochs
elif args.dataset == "robotcar":
args.envs = ["sun", "night", "overcast"]
args.epochs = [3] if args.epochs is None else args.epochs
args.epochs = [1] if args.epochs is None else args.epochs
n_env = len(args.envs)
if len(args.epochs) == 1:
args.epochs = args.epochs * n_env
else:
assert args.method != 'joint' and len(args.epochs) == n_env
out_dir = Path(args.out_dir)
all_env_regex = '(' + '|'.join(args.envs) + ')'
save_path = create_dir(out_dir / 'train') / 'model.pth'
for i, (epoch, env) in enumerate(zip(args.epochs, args.envs)):
if not args.skip_train:
train_args = ['--task', 'train-joint' if args.method == 'joint' else 'train-seq']
train_args += ['--dataset', args.dataset]
train_args += ['--include', all_env_regex if args.method == 'joint' else env]
train_args += ['--epoch', str(epoch)]
# load model saved from previous runx
if i > 0:
train_args += ['--load', str(save_path)]
train_args += ['--save', str(save_path)]
# weights loading for lifelong methods
if args.method not in ['finetune', 'joint']:
train_args += ['--ll-method', args.method]
train_args += ['--ll-weight-dir', str(create_dir(out_dir / 'll-weights'))]
if i > 0:
train_args += ['--ll-weight-load'] + args.envs[:i]
run(train_args + extra_args)
if args.method == 'joint':
save_path = save_path.parent / (save_path.name + f'.epoch{epoch - 1}')
else:
save_path = save_path.parent / (save_path.name + (f".{env}.{epoch - 1}" if epoch > 1 else f".{env}"))
if not args.skip_eval:
eval_args = ['--task', 'eval', '--dataset', args.dataset, '--include', all_env_regex, '--load', str(save_path)]
run(eval_args + extra_args)
def create_dir(directory: Path):
directory.mkdir(parents=True, exist_ok=True)
return directory
if __name__ == '__main__':
parser = configargparse.ArgParser()
# meta
parser.add_argument('--out-dir', type=str, default="./run/")
parser.add_argument('--skip-eval', action='store_true')
parser.add_argument('--skip-train', action='store_true')
# launch
parser.add_argument("--dataset", type=str, default='tartanair',
choices=['tartanair', 'nordland', 'robotcar'], help="Dataset to use")
parser.add_argument('--envs', type=str, nargs='+')
parser.add_argument('--epochs', type=int, nargs='+')
parser.add_argument('--method', type=str, required=True,
choices=['finetune', 'si', 'ewc', 'kd', 'rkd', 'mas', 'rmas', 'airloop', 'joint'])
parserd_args, unknown_args = parser.parse_known_args()
main(parserd_args, unknown_args)