-
Notifications
You must be signed in to change notification settings - Fork 112
/
config.py
125 lines (104 loc) · 5.41 KB
/
config.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
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
import argparse
arg_lists = []
parser = argparse.ArgumentParser()
def add_argument_group(name):
arg = parser.add_argument_group(name)
arg_lists.append(arg)
return arg
def str2bool(v):
return v.lower() in ('true', '1')
logging_arg = add_argument_group('Logging')
logging_arg.add_argument('--out_dir', type=str, default='outputs')
trainer_arg = add_argument_group('Trainer')
trainer_arg.add_argument('--trainer', type=str, default='HardestContrastiveLossTrainer')
trainer_arg.add_argument('--save_freq_epoch', type=int, default=1)
trainer_arg.add_argument('--batch_size', type=int, default=4)
trainer_arg.add_argument('--val_batch_size', type=int, default=1)
# Hard negative mining
trainer_arg.add_argument('--use_hard_negative', type=str2bool, default=True)
trainer_arg.add_argument('--hard_negative_sample_ratio', type=int, default=0.05)
trainer_arg.add_argument('--hard_negative_max_num', type=int, default=3000)
trainer_arg.add_argument('--num_pos_per_batch', type=int, default=1024)
trainer_arg.add_argument('--num_hn_samples_per_batch', type=int, default=256)
# Metric learning loss
trainer_arg.add_argument('--neg_thresh', type=float, default=1.4)
trainer_arg.add_argument('--pos_thresh', type=float, default=0.1)
trainer_arg.add_argument('--neg_weight', type=float, default=1)
# Data augmentation
trainer_arg.add_argument('--use_random_scale', type=str2bool, default=False)
trainer_arg.add_argument('--min_scale', type=float, default=0.8)
trainer_arg.add_argument('--max_scale', type=float, default=1.2)
trainer_arg.add_argument('--use_random_rotation', type=str2bool, default=True)
trainer_arg.add_argument('--rotation_range', type=float, default=360)
# Data loader configs
trainer_arg.add_argument('--train_phase', type=str, default="train")
trainer_arg.add_argument('--val_phase', type=str, default="val")
trainer_arg.add_argument('--test_phase', type=str, default="test")
trainer_arg.add_argument('--stat_freq', type=int, default=40)
trainer_arg.add_argument('--test_valid', type=str2bool, default=True)
trainer_arg.add_argument('--val_max_iter', type=int, default=400)
trainer_arg.add_argument('--val_epoch_freq', type=int, default=1)
trainer_arg.add_argument(
'--positive_pair_search_voxel_size_multiplier', type=float, default=1.5)
trainer_arg.add_argument('--hit_ratio_thresh', type=float, default=0.1)
# Triplets
trainer_arg.add_argument('--triplet_num_pos', type=int, default=256)
trainer_arg.add_argument('--triplet_num_hn', type=int, default=512)
trainer_arg.add_argument('--triplet_num_rand', type=int, default=1024)
# dNetwork specific configurations
net_arg = add_argument_group('Network')
net_arg.add_argument('--model', type=str, default='ResUNetBN2C')
net_arg.add_argument('--model_n_out', type=int, default=32, help='Feature dimension')
net_arg.add_argument('--conv1_kernel_size', type=int, default=5)
net_arg.add_argument('--normalize_feature', type=str2bool, default=True)
net_arg.add_argument('--dist_type', type=str, default='L2')
net_arg.add_argument('--best_val_metric', type=str, default='feat_match_ratio')
# Optimizer arguments
opt_arg = add_argument_group('Optimizer')
opt_arg.add_argument('--optimizer', type=str, default='SGD')
opt_arg.add_argument('--max_epoch', type=int, default=100)
opt_arg.add_argument('--lr', type=float, default=1e-1)
opt_arg.add_argument('--momentum', type=float, default=0.8)
opt_arg.add_argument('--sgd_momentum', type=float, default=0.9)
opt_arg.add_argument('--sgd_dampening', type=float, default=0.1)
opt_arg.add_argument('--adam_beta1', type=float, default=0.9)
opt_arg.add_argument('--adam_beta2', type=float, default=0.999)
opt_arg.add_argument('--weight_decay', type=float, default=1e-4)
opt_arg.add_argument('--iter_size', type=int, default=1, help='accumulate gradient')
opt_arg.add_argument('--bn_momentum', type=float, default=0.05)
opt_arg.add_argument('--exp_gamma', type=float, default=0.99)
opt_arg.add_argument('--scheduler', type=str, default='ExpLR')
opt_arg.add_argument(
'--icp_cache_path', type=str, default="/home/chrischoy/datasets/FCGF/kitti/icp/")
misc_arg = add_argument_group('Misc')
misc_arg.add_argument('--use_gpu', type=str2bool, default=True)
misc_arg.add_argument('--weights', type=str, default=None)
misc_arg.add_argument('--weights_dir', type=str, default=None)
misc_arg.add_argument('--resume', type=str, default=None)
misc_arg.add_argument('--resume_dir', type=str, default=None)
misc_arg.add_argument('--train_num_thread', type=int, default=2)
misc_arg.add_argument('--val_num_thread', type=int, default=1)
misc_arg.add_argument('--test_num_thread', type=int, default=2)
misc_arg.add_argument('--fast_validation', type=str2bool, default=False)
misc_arg.add_argument(
'--nn_max_n',
type=int,
default=500,
help='The maximum number of features to find nearest neighbors in batch')
# Dataset specific configurations
data_arg = add_argument_group('Data')
data_arg.add_argument('--dataset', type=str, default='ThreeDMatchPairDataset')
data_arg.add_argument('--voxel_size', type=float, default=0.025)
data_arg.add_argument(
'--threed_match_dir', type=str, default="/home/chrischoy/datasets/FCGF/threedmatch")
data_arg.add_argument(
'--kitti_root', type=str, default="/home/chrischoy/datasets/FCGF/kitti/")
data_arg.add_argument(
'--kitti_max_time_diff',
type=int,
default=3,
help='max time difference between pairs (non inclusive)')
data_arg.add_argument('--kitti_date', type=str, default='2011_09_26')
def get_config():
args = parser.parse_args()
return args