-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
156 lines (147 loc) · 4.3 KB
/
train.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
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
import argparse
from time import time
import logging
import numpy as np
import torch
import random
import os
from PC_lib import PCTrainer
def get_parser():
parser = argparse.ArgumentParser(description="Train motion net")
parser.add_argument(
"--test",
action="store_true",
help="indicating whether to only do evaluation",
)
parser.add_argument(
"--continue_train",
action="store_true",
help="indicating whether to only do evaluation",
)
parser.add_argument(
"--continue_model",
default=None,
metavar="FILE",
help="the model filefor continue training",
)
parser.add_argument(
"--train_path",
default=None,
metavar="FILE",
help="hdf5 file which contains the train data",
)
parser.add_argument(
"--test_path",
required=True,
metavar="FILE",
help="hdf5 file which contains the test data",
)
parser.add_argument(
"--inference_model",
default=None,
metavar="FILE",
help="the inference file when test is True",
)
parser.add_argument(
"--output_dir",
required=True,
metavar="DIR",
help="hdf5 file which contains the test data",
)
parser.add_argument(
"--max_K",
default=10,
type=int,
help="indicatet the max number for the segmentation",
)
parser.add_argument(
"--category_number",
default=4,
type=int,
help="indicate the number of part categories",
)
# Some default hyperparameters
parser.add_argument(
"--device",
default="cuda:0",
help="choose from cuda or cpu",
)
parser.add_argument(
"--max_epochs",
type=int,
default=600,
)
parser.add_argument(
"--lr",
type=float,
default=0.001,
)
parser.add_argument(
"--batch_size",
type=int,
default=32,
)
parser.add_argument(
"--num_workers",
type=int,
default=8,
)
parser.add_argument(
"--num_points",
default=4096,
help="Number of points used to sample the input data"
)
parser.add_argument(
"--random_seed",
default=42,
)
parser.add_argument(
"--num_channels",
default=6,
help="Use xyz, RGB, and normal"
)
return parser
if __name__ == "__main__":
start = time()
args = get_parser().parse_args()
logging.basicConfig(level=logging.DEBUG)
log = logging.getLogger('train')
log.info("Arguments: " + str(args))
# Setup some default value
args.data_path = {"train": args.train_path, "test": args.test_path}
args.save_frequency = 50
args.log_frequency = 1
args.loss_weight = {
"loss_category": 1.0,
"loss_instance": 1.0,
"loss_mtype": 1.0,
"loss_maxis": 1.0,
"loss_morigin": 1.0,
}
# Make the training deterministic
seed = int(args.random_seed)
np.random.seed(seed)
torch.set_rng_state(torch.manual_seed(seed).get_state())
random.seed(seed)
torch.set_deterministic(True)
torch.backends.cudnn.deterministic = True
os.environ["CUBLAS_WORKSPACE_CONFIG"] = ":4096:8"
trainer = PCTrainer(args, args.max_K, args.category_number, args.num_channels)
if not args.test:
if not args.continue_train:
if args.continue_model == None:
log.info(f'Train on {args.train_path}, validate on {args.test_path}')
trainer.train()
else:
log.info(f'Train on {args.train_path}, validate on {args.test_path} with pretrined model {args.continue_model}')
checkpoint = torch.load(args.continue_model, map_location=trainer.device)
trainer.model.load_state_dict(checkpoint["model_state_dict"])
trainer.train()
else:
log.info(f'Continue training with {args.continue_model} on {args.train_path}, validate on {args.test_path}')
trainer.resume_train(model_path=args.continue_model)
else:
log.info(f'Test on {args.test_path} with inference model {args.inference_model}')
trainer.test(inference_model=args.inference_model)
stop = time()
log.info(f"Total time: {stop-start}")