/
train.py
executable file
·171 lines (125 loc) · 6.29 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
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
import os
import sys
import copy
import math
import wandb
import numpy as np
from pathlib import Path
from collections import OrderedDict
import torch
import torch.nn as nn
import torch.optim as optim
from utils.dataset import SegDataset, create_loader
from utils.evaluation import eval_net
from utils.loss import muti_ce_loss, multi_mse_loss
from utils.utils import *
# config_path = sys.argv[1]
config_path = sys.argv[1]
opt = load_yaml(config_path)
device = opt['device']
setup_seed(opt['seed'])
train_logger, val_logger = get_logger(opt['path']['logs'])
for phase, dataset_opt in opt['datasets'].items():
if phase == 'train':
train_set = SegDataset(dataset_opt, opt['dataset'], opt['task'], dim=opt['hog_decoder']['out_dim'], is_train=True)
train_size = int(
math.ceil(len(train_set) / dataset_opt['batch_size']))
train_logger.info('Number of train images: {:,d}, iters: {:,d}'.format(
len(train_set), train_size))
total_iters = int(opt['train']['niters'])
total_epochs = int(math.ceil(total_iters / train_size))
train_logger.info('Total epochs needed: {:d} for iters {:,d}'.format(
total_epochs, total_iters))
train_loader = create_loader(
train_set, opt=dataset_opt, is_train=True)
elif phase == 'valid':
val_set = SegDataset(dataset_opt, opt['dataset'], opt['task'], is_train=False)
val_loader = create_loader(
val_set, opt=dataset_opt, is_train=False)
train_logger.info('Number of validation images in [{:s}]: {:d}'.format(dataset_opt['name'],
len(val_set)))
else:
raise NotImplementedError(
'Phase [{:s}] is not recognized.'.format(phase))
assert train_loader is not None
wandb.login()
wandb.init(project=opt['wandb']['project'], entity=opt['wandb']['entity'], name=opt['wandb']['name'])
seg_net, hog_decoder = define_networks(opt)
optim_seg = optim.Adam(seg_net.parameters(), lr=opt['train']['lr_seg_net'], betas=(opt['train']['b1_seg_net'], opt['train']['b2_seg_net']), eps=1e-08, weight_decay=1e-7)
sch_seg= optim.lr_scheduler.MultiStepLR(optim_seg, opt['train']['lr_steps'], opt['train']['lr_gamma'])
if hog_decoder is not None:
optim_hog = optim.Adam(hog_decoder.parameters(), lr=opt['train']['lr_hog_dec'], betas=(opt['train']['b1_hog_dec'], opt['train']['b2_hog_dec']), eps=1e-08, weight_decay=1e-7)
sch_hog = optim.lr_scheduler.MultiStepLR(optim_hog, opt['train']['lr_steps'], opt['train']['lr_gamma'])
start_epoch = opt['train']['start_epoch']
ite_num = train_size * start_epoch
best_miou = 0
best_ite = 0
best_seg_net = None
best_hog_decoder = None
train_logger.info('Start training from epoch: {:d}, iter: {:d}'.format(
start_epoch, ite_num))
log_dict = OrderedDict()
ckpt_dir = opt['path']['checkpoints']['models']
mkdirs(ckpt_dir)
for epoch in range(start_epoch, total_epochs):
for _, train_data in enumerate(train_loader):
ite_num = ite_num + 1
inputs, labels, hog_true = train_data['image'].to(device), train_data['mask'].to(device), train_data['hog_f'].to(device)
optim_seg.zero_grad()
optim_hog.zero_grad()
d0, d1, d2, d3, d4, d5, d6, hx1d, hx2d, hx3d, hx4d, hx5d, hx6 = seg_net(inputs)
loss_ce = muti_ce_loss(d0, d1, d2, d3, d4, d5, d6, labels)
loss_total = loss_ce
if hog_decoder is not None:
hog1, hog2, hog3, hog4, hog5, hog6 = hog_decoder(hx1d, hx2d, hx3d, hx4d, hx5d, hx6)
loss_mse = multi_mse_loss(hog1, hog2, hog3, hog4, hog5, hog6, hog_true)
loss_total = loss_ce + loss_mse
loss_total.backward()
optim_seg.step()
sch_seg.step()
if hog_decoder is not None:
optim_hog.step()
sch_hog.step()
log_dict['CE_loss'] = loss_ce.item()
log_dict['MSE_loss'] = loss_mse.item()
log_dict['LR_seg_net'] = sch_seg.get_last_lr()[0]
log_dict['LR_hog_net'] = sch_hog.get_last_lr()[0]
wandb.log({"Train/CE_loss":loss_ce.item()}, step=ite_num)
wandb.log({"Train/MSE_loss":loss_mse.item()}, step=ite_num)
wandb.log({"Train/LR_seg_net":sch_seg.get_last_lr()[0]}, step=ite_num)
wandb.log({"Train/LR_hog_net":sch_hog.get_last_lr()[0]}, step=ite_num)
if ite_num % opt['train']['print_freq'] == 0:
if hog_decoder is not None:
message = '<epoch:{:3d}, iter:{:8,d}, lr_seg:{:.3e}, lr_hog:{:.3e}> '.format(
epoch, ite_num, sch_seg.get_last_lr()[0], sch_hog.get_last_lr()[0])
else:
message = '<epoch:{:3d}, iter:{:8,d}, lr_seg:{:.3e}> '.format(
epoch, ite_num, sch_seg.get_last_lr()[0])
for k, v in log_dict.items():
message += '{:s}: {:.4e} '.format(k, v)
train_logger.info(message)
if ite_num % opt['train']['save_step'] == 0:
torch.save(seg_net.state_dict(), os.path.join(ckpt_dir, f'seg_net_{ite_num}.pth'))
if hog_decoder is not None:
torch.save(hog_decoder.state_dict(), os.path.join(ckpt_dir, f'hog_dec_{ite_num}.pth'))
train_logger.info('Saved Checkpoints.')
if ite_num % opt['train']['val_freq'] == 0:
ce_val_loss, miou = eval_net(opt, seg_net, val_loader, device, ite_num)
if miou.item() > best_miou:
best_miou = miou.item()
best_ite = ite_num
best_seg = copy.deepcopy(seg_net)
best_hog = copy.deepcopy(hog_decoder)
wandb.log({"Valid/CE_loss": ce_val_loss,
"Metric/mIOU": miou.item()},
step=ite_num)
val_logger.info('<epoch:{:3d}, iter:{:8,d}> mIOU: {:.4e} Val_loss_ce: {:.4e}'.format(
epoch, ite_num, miou.item(), ce_val_loss))
train_logger.info(f'Iteration: {ite_num}\t mIOU: {miou.item()}\t Val_loss_ce: {ce_val_loss}')
if total_iters <= ite_num:
torch.save(seg_net.state_dict(), os.path.join(ckpt_dir, f'best_seg_net_{ite_num}.pth'))
if hog_decoder is not None:
torch.save(hog_decoder.state_dict(), os.path.join(ckpt_dir, f'best_hog_dec_{ite_num}.pth'))
break
if total_iters <= ite_num:
break