-
Notifications
You must be signed in to change notification settings - Fork 0
/
pre_train.py
126 lines (105 loc) · 5.14 KB
/
pre_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
import torch
from time import time
import numpy as np
import os
from src.model_baseline import compile_model_lss
from src.data_pretrain import compile_data
from src.tools import SimpleLoss, get_val_info
def train(args):
max_grad_norm = 5.0
grid_conf = {'xbound': args.xbound, 'ybound': args.ybound,
'zbound': args.zbound,'dbound': args.dbound,}
data_aug_conf = {
'resize_lim': args.resize_lim,
'final_dim': args.final_dim,
'rot_lim': args.rot_lim,
'H': args.H, 'W': args.W,
'rand_flip': args.rand_flip,
'bot_pct_lim': args.bot_pct_lim,
'cams': ['CAM_FRONT_LEFT', 'CAM_FRONT', 'CAM_FRONT_RIGHT',
'CAM_BACK_LEFT', 'CAM_BACK', 'CAM_BACK_RIGHT'],
'Ncams': args.ncams,
}
trainloader, valloader = compile_data(args.version, args.dataroot, data_aug_conf=data_aug_conf,
grid_conf=grid_conf, bsz=args.bsize, nworkers=args.nworkers,
parser_name='segmentationdata')
if not os.path.exists(args.logdir):
os.mkdir(args.logdir)
device = torch.device('cpu') if args.gpuid < 0 else torch.device(f'cuda:{args.gpuid}')
model = compile_model_lss(args.bsize, grid_conf, data_aug_conf, outC=args.seg_classes)
if args.checkpoint:
print('loading', args.checkpoint)
model.load_state_dict(torch.load(args.checkpoint), strict=True)
model.to(device)
opt = torch.optim.Adam(model.parameters(), lr=args.lr, weight_decay=args.wdecay)
loss_fn = SimpleLoss().cuda(args.gpuid)
counter = 0
for epoch in range(args.nepochs):
print('--------------Epoch: {}--------------'.format(epoch))
np.random.seed()
model.train()
for batchi, (imgs, rots, trans, intrins, post_rots, post_trans, binimgs) in enumerate(trainloader):
t0 = time()
opt.zero_grad()
preds = model(imgs.to(device),
rots.to(device),
trans.to(device),
intrins.to(device),
post_rots.to(device),
post_trans.to(device),)
binimgs = binimgs.to(device)
loss = loss_fn(preds, binimgs)
loss.backward()
torch.nn.utils.clip_grad_norm_(model.parameters(), max_grad_norm)
opt.step()
counter += 1
t1 = time()
if counter % 200 == 0:
print('Counter{} Train_Loss: {}'.format(counter, loss.item()))
# val_info
iou_info, val_loss = get_val_info(model, valloader, loss_fn, device)
iou_info = str(iou_info)
print(iou_info)
print('val_loss: {}'.format(val_loss))
# Log the val info
results_txt = './pretrain-result.txt'
with open(results_txt, "a") as f:
f.write('epoch{}'.format(epoch) + iou_info + '\n' + 'val_loss: ' + str(val_loss) + "\n\n")
# Save the weight
mname = os.path.join(args.logdir, "model{}.pt".format(epoch))
print('saving', mname)
torch.save(model.state_dict(), mname)
model.train()
def parse_args():
import argparse
parser = argparse.ArgumentParser(description="pytorch pre-training")
# General Setting
parser.add_argument("--version", default='trainval', help='[trainval, mini]')
parser.add_argument("--dataroot", default="/path/to/the/dataset/")
parser.add_argument("--nepochs", default=10000, type=int)
parser.add_argument("--gpuid", default=1, type=int)
parser.add_argument("--logdir", default='./pretrain_result/', help='path for the log file')
parser.add_argument("--bsize", default=6, type=int)
parser.add_argument("--nworkers", default=10, type=int)
parser.add_argument('--lr', default=1e-3, type=float, help='initial learning rate')
parser.add_argument('--wdecay', default=1e-7, type=float, help='weight decay')
parser.add_argument('--checkpoint', default='')
parser.add_argument('--examplef', default='./examples', help='file for viz images')
parser.add_argument('--seg_classes', default=4, help='number of class in segmentation')
parser.add_argument('--xbound', default=[-50.0, 50.0, 0.5], help='grid configuration')
parser.add_argument('--ybound', default=[-50.0, 50.0, 0.5], help='grid configuration')
parser.add_argument('--zbound', default=[-10.0, 10.0, 20.0], help='grid configuration')
parser.add_argument('--dbound', default=[4.0, 45.0, 1.0], help='grid configuration')
parser.add_argument('--H', default=900, type=int)
parser.add_argument('--W', default=1600, type=int)
parser.add_argument('--resize_lim', default=(0.193, 0.225))
parser.add_argument('--final_dim', default=(128, 352))
parser.add_argument('--bot_pct_lim', default=(0.0, 0.22))
parser.add_argument('--rot_lim', default=(-5.4, 5.4))
parser.add_argument('--rand_flip', default=True, type=bool)
parser.add_argument('--ncams', default=6, type=int)
args = parser.parse_args()
return args
if __name__ == '__main__':
args = parse_args()
train(args)