-
Notifications
You must be signed in to change notification settings - Fork 6
/
train.py
203 lines (177 loc) · 7.76 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
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
# -*- coding: utf-8 -*-
"""
@author: caigentan@AnHui University
@software: PyCharm
@file: train.py
@time: 2021/11/25 10:11
"""
import os
import torch
import torch.nn.functional as F
import sys
sys.path.append('./models')
import numpy as np
from datetime import datetime
from models.hrt import HighResolutionTransformer
from torchvision.utils import make_grid
from data import get_loader, test_dataset
from utils import clip_gradient, adjust_lr
from tensorboardX import SummaryWriter
from tools.pytorch_utils import Save_Handle, AverageMeter
import tools.log_utils as log_utils
import torch.backends.cudnn as cudnn
from options import opt
import yaml
save_list = Save_Handle(max_num=1)
path = "./config/hrt_base.yaml"
config = yaml.load(open(path, 'r'), yaml.SafeLoader)['MODEL']['HRT']
# set the device for training
model_name = opt.model_name
image_root = opt.rgb_root
gt_root = opt.gt_root
depth_root = opt.depth_root
edge_root = opt.edge_root
test_image_root = opt.test_rgb_root
test_gt_root = opt.test_gt_root
test_depth_root = opt.test_depth_root
save_path = opt.save_path
# set the path
if not os.path.exists(save_path):
os.makedirs(save_path)
time_str = datetime.strftime(datetime.now(),"%m%d-%H%M%S")
logger = log_utils.get_logger(save_path + 'train-{}-{:s}.log'.format(model_name, time_str))
log_utils.print_config(vars(opt), logger)
if opt.gpu_id == '0':
os.environ["CUDA_VISIBLE_DEVICES"] = "0"
logger.info('using gpu 0')
elif opt.gpu_id == '1':
os.environ["CUDA_VISIBLE_DEVICES"] = "1"
logger.info('using gpu 1')
cudnn.benchmark = True
model = HighResolutionTransformer(config, 1000)
model.cuda()
params = model.parameters()
optimizer = torch.optim.Adam(params, opt.lr)
num_parms = 0
for p in model.parameters():
num_parms += p.numel()
logger.info("Total Parameters (For Reference): {}".format(num_parms))
start_epoch = 0
if (opt.hr_load is not None):
model.init_weights(opt.hr_load, opt.cnn_load)
logger.info('loading pretrained model from ' + opt.hr_load)
logger.info('loading pretrained model from ' + opt.cnn_load)
elif opt.resume:
logger.info('loading pretrained model from last stop ' + opt.resume)
suf = opt.resume.rsplit('.', 1)[-1]
if suf == "tar":
checkpoint = torch.load(opt.resume, torch.device('cuda'))
model.load_state_dict(checkpoint['model_state_dict'])
optimizer.load_state_dict(checkpoint['optimizer_state_dict'])
start_epoch = checkpoint['epoch'] + 1
logger.info('model load successfully, optimizer load successfully!')
def structure_loss(pred, mask):
weit = 1+5*torch.abs(F.avg_pool2d(mask, kernel_size=31, stride=1, padding=15)-mask)
wbce = F.binary_cross_entropy_with_logits(pred, mask, reduce='none')
wbce = (weit*wbce).sum(dim=(2,3))/weit.sum(dim=(2,3))
pred = torch.sigmoid(pred)
inter = ((pred*mask)*weit).sum(dim=(2,3))
union = ((pred+mask)*weit).sum(dim=(2,3))
wiou = 1-(inter+1)/(union-inter+1)
return (wbce+wiou).mean()
step = 0
writer = SummaryWriter(save_path + 'summary')
best_mae = 1
best_epoch = 0
# load data
logger.info('load data...')
train_loader = get_loader(image_root, gt_root,depth_root, edge_root, batchsize=opt.batchsize, trainsize=opt.trainsize)
test_loader = test_dataset(test_image_root, test_gt_root,test_depth_root, opt.trainsize)
total_step = len(train_loader)
# train function
def train(train_loader, model, optimizer, epoch, save_path):
global step
model.train()
loss_all = 0
epoch_step = 0
try:
for i, (images, gts, depth,edge) in enumerate(train_loader, start=1):
optimizer.zero_grad()
images = images.cuda()
gts = gts.cuda()
depth = depth.cuda()
s = model(images,depth)
loss = structure_loss(s, gts)
loss.backward()
clip_gradient(optimizer, opt.clip)
optimizer.step()
step += 1
epoch_step += 1
loss_all += loss.data
memory_used = torch.cuda.max_memory_allocated() / (1024.0 * 1024.0)
if i % 100 == 0 or i == total_step or i == 1:
logger.info(
'#TRAIN#:Epoch [{:03d}/{:03d}], Step [{:04d}/{:04d}], LR:{:.7f}, sal_loss:{:4f} ||Mem_use:{:.0f}MB'.
format(epoch, opt.epoch, i, total_step, optimizer.state_dict()['param_groups'][0]['lr'], loss.data, memory_used))
writer.add_scalar('Loss', loss.data, global_step=step)
grid_image = make_grid(images[0].clone().cpu().data, 1, normalize=True)
writer.add_image('RGB', grid_image, step)
grid_image = make_grid(gts[0].clone().cpu().data, 1, normalize=True)
writer.add_image('Ground_truth', grid_image, step)
res = s[0].clone()
res = res.sigmoid().data.cpu().numpy().squeeze()
res = (res - res.min()) / (res.max() - res.min() + 1e-8)
writer.add_image('res', torch.tensor(res), step, dataformats='HW')
loss_all /= epoch_step
logger.info('#TRAIN#:Epoch [{:03d}/{:03d}],Loss_AVG: {:.4f}'.format(epoch, opt.epoch, loss_all))
writer.add_scalar('Loss-epoch', loss_all, global_step=epoch)
if (epoch) % 5 == 0:
torch.save(model.state_dict(), save_path + '{}_epoch_{}.pth'.format(model_name,epoch))
temp_save_path = save_path + "{}_ckpt.tar".format(epoch)
torch.save({
'epoch': epoch,
'optimizer_state_dict': optimizer.state_dict(),
'model_state_dict': model.state_dict()
}, temp_save_path)
save_list.append(temp_save_path)
except KeyboardInterrupt:
logger.info('Keyboard Interrupt: save model and exit.')
if not os.path.exists(save_path):
os.makedirs(save_path)
torch.save(model.state_dict(), save_path + '{}_epoch_{}.pth'.format(model_name,epoch + 1))
logger.info('save checkpoints successfully!')
raise
def test(test_loader, model, epoch, save_path):
global best_mae, best_epoch
model.eval()
with torch.no_grad():
mae_sum = 0
for i in range(test_loader.size):
image, gt, depth, name, img_for_post = test_loader.load_data()
gt = np.asarray(gt, np.float32)
gt /= (gt.max() + 1e-8)
image = image.cuda()
depth = depth.cuda()
res = model(image,depth)
res = F.upsample(res, size=gt.shape, mode='bilinear', align_corners=False)
res = res.sigmoid().data.cpu().numpy().squeeze()
res = (res - res.min()) / (res.max() - res.min() + 1e-8)
mae_sum += np.sum(np.abs(res - gt)) * 1.0 / (gt.shape[0] * gt.shape[1])
mae = mae_sum / test_loader.size
writer.add_scalar('MAE', torch.tensor(mae), global_step=epoch)
logger.info('Epoch: {} MAE: {} #### bestMAE: {} bestEpoch: {}'.format(epoch, mae, best_mae, best_epoch))
if epoch == 1:
best_mae = mae
else:
if mae < best_mae:
best_mae = mae
best_epoch = epoch
torch.save(model.state_dict(), save_path + '{}_epoch_best.pth'.format(model_name))
logger.info('#TEST#:Epoch:{} MAE:{} bestEpoch:{} bestMAE:{}'.format(epoch, mae, best_epoch, best_mae))
if __name__ == '__main__':
logger.info("Start train...")
for epoch in range(start_epoch, opt.epoch+1):
cur_lr = adjust_lr(optimizer, opt.lr, epoch, opt.decay_rate, opt.decay_epoch)
writer.add_scalar('learning_rate', cur_lr, global_step=epoch)
train(train_loader, model, optimizer, epoch, save_path)
test(test_loader, model, epoch, save_path)