This repository was archived by the owner on Sep 22, 2024. It is now read-only.
forked from qinnzou/DeepCrack
-
Notifications
You must be signed in to change notification settings - Fork 0
Expand file tree
/
Copy pathtrain.py
More file actions
201 lines (167 loc) · 10.2 KB
/
train.py
File metadata and controls
201 lines (167 loc) · 10.2 KB
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
from data.augmentation import augCompose, RandomBlur, RandomColorJitter
from data.dataset import readIndex, dataReadPip, loadedDataset
from tqdm import tqdm
from model.deepcrack import DeepCrack
from trainer import DeepCrackTrainer
from config import Config as cfg
import numpy as np
import torch
import os
import cv2
import sys
os.environ["CUDA_VISIBLE_DEVICES"] = cfg.gpu_id
def main():
# ----------------------- dataset ----------------------- #
data_augment_op = augCompose(transforms=[[RandomColorJitter, 0.5], [RandomBlur, 0.2]])
train_pipline = dataReadPip(transforms=data_augment_op)
test_pipline = dataReadPip(transforms=None)
train_dataset = loadedDataset(readIndex(cfg.train_data_path, shuffle=True), preprocess=train_pipline)
test_dataset = loadedDataset(readIndex(cfg.test_data_path), preprocess=test_pipline)
train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=cfg.train_batch_size,
shuffle=True, num_workers=4, drop_last=True)
val_loader = torch.utils.data.DataLoader(test_dataset, batch_size=cfg.val_batch_size,
shuffle=False, num_workers=4, drop_last=True)
# -------------------- build trainer --------------------- #
device = torch.device("cuda")
num_gpu = torch.cuda.device_count()
model = DeepCrack()
model = torch.nn.DataParallel(model, device_ids=range(num_gpu))
model.to(device)
trainer = DeepCrackTrainer(model).to(device)
if cfg.pretrained_model:
pretrained_dict = trainer.saver.load(cfg.pretrained_model, multi_gpu=True)
model_dict = model.state_dict()
# pretrained_dict = {k: v for k, v in pretrained_dict.items() if k in model_dict}
# model_dict.update(pretrained_dict)
pretrained_dict = {key.replace("module.", ""): value for key, value in pretrained_dict.items()}
model_dict.update(pretrained_dict)
model.load_state_dict(model_dict)
trainer.vis.log('load checkpoint: %s' % cfg.pretrained_model, 'train info')
try:
for epoch in range(1, cfg.epoch):
trainer.vis.log('Start Epoch %d ...' % epoch, 'train info')
model.train()
# --------------------- training ------------------- #
bar = tqdm(enumerate(train_loader), total=len(train_loader))
bar.set_description('Epoch %d --- Training --- :' % epoch)
for idx, (img, lab) in bar:
data, target = img.type(torch.cuda.FloatTensor).to(device), lab.type(torch.cuda.FloatTensor).to(device)
pred = trainer.train_op(data, target)
if idx % cfg.vis_train_loss_every == 0:
trainer.vis.log(trainer.log_loss, 'train_loss')
trainer.vis.plot_many({
'train_total_loss': trainer.log_loss['total_loss'],
'train_output_loss': trainer.log_loss['output_loss'],
'train_fuse5_loss': trainer.log_loss['fuse5_loss'],
'train_fuse4_loss': trainer.log_loss['fuse4_loss'],
'train_fuse3_loss': trainer.log_loss['fuse3_loss'],
'train_fuse2_loss': trainer.log_loss['fuse2_loss'],
'train_fuse1_loss': trainer.log_loss['fuse1_loss'],
})
if idx % cfg.vis_train_acc_every == 0:
trainer.acc_op(pred[0], target)
trainer.vis.log(trainer.log_acc, 'train_acc')
trainer.vis.plot_many({
'train_mask_acc': trainer.log_acc['mask_acc'],
'train_mask_pos_acc': trainer.log_acc['mask_pos_acc'],
'train_mask_neg_acc': trainer.log_acc['mask_neg_acc'],
})
if idx % cfg.vis_train_img_every == 0:
trainer.vis.img_many({
'train_img': data.cpu(),
'train_output': torch.sigmoid(pred[0].contiguous().cpu()),
'train_lab': target.unsqueeze(1).cpu(),
'train_fuse5': torch.sigmoid(pred[1].contiguous().cpu()),
'train_fuse4': torch.sigmoid(pred[2].contiguous().cpu()),
'train_fuse3': torch.sigmoid(pred[3].contiguous().cpu()),
'train_fuse2': torch.sigmoid(pred[4].contiguous().cpu()),
'train_fuse1': torch.sigmoid(pred[5].contiguous().cpu()),
})
if idx % cfg.val_every == 0:
trainer.vis.log('Start Val %d ....' % idx, 'train info')
# -------------------- val ------------------- #
model.eval()
val_loss = {
'eval_total_loss': 0,
'eval_output_loss': 0,
'eval_fuse5_loss': 0,
'eval_fuse4_loss': 0,
'eval_fuse3_loss': 0,
'eval_fuse2_loss': 0,
'eval_fuse1_loss': 0,
}
val_acc = {
'mask_acc': 0,
'mask_pos_acc': 0,
'mask_neg_acc': 0,
}
bar.set_description('Epoch %d --- Evaluation --- :' % epoch)
with torch.no_grad():
for idx, (img, lab) in enumerate(val_loader, start=1):
val_data, val_target = img.type(torch.cuda.FloatTensor).to(device), lab.type(
torch.cuda.FloatTensor).to(device)
val_pred = trainer.val_op(val_data, val_target)
trainer.acc_op(val_pred[0], val_target)
val_loss['eval_total_loss'] += trainer.log_loss['total_loss']
val_loss['eval_output_loss'] += trainer.log_loss['output_loss']
val_loss['eval_fuse5_loss'] += trainer.log_loss['fuse5_loss']
val_loss['eval_fuse4_loss'] += trainer.log_loss['fuse4_loss']
val_loss['eval_fuse3_loss'] += trainer.log_loss['fuse3_loss']
val_loss['eval_fuse2_loss'] += trainer.log_loss['fuse2_loss']
val_loss['eval_fuse1_loss'] += trainer.log_loss['fuse1_loss']
val_acc['mask_acc'] += trainer.log_acc['mask_acc']
val_acc['mask_pos_acc'] += trainer.log_acc['mask_pos_acc']
val_acc['mask_neg_acc'] += trainer.log_acc['mask_neg_acc']
else:
trainer.vis.img_many({
'eval_img': val_data.cpu(),
'eval_output': torch.sigmoid(val_pred[0].contiguous().cpu()),
'eval_lab': val_target.unsqueeze(1).cpu(),
'eval_fuse5': torch.sigmoid(val_pred[1].contiguous().cpu()),
'eval_fuse4': torch.sigmoid(val_pred[2].contiguous().cpu()),
'eval_fuse3': torch.sigmoid(val_pred[3].contiguous().cpu()),
'eval_fuse2': torch.sigmoid(val_pred[4].contiguous().cpu()),
'eval_fuse1': torch.sigmoid(val_pred[5].contiguous().cpu()),
})
trainer.vis.plot_many({
'eval_total_loss': val_loss['eval_total_loss'] / idx,
'eval_output_loss': val_loss['eval_output_loss'] / idx,
'eval_fuse5_loss': val_loss['eval_fuse5_loss'] / idx,
'eval_fuse4_loss': val_loss['eval_fuse4_loss'] / idx,
'eval_fuse3_loss': val_loss['eval_fuse3_loss'] / idx,
'eval_fuse2_loss': val_loss['eval_fuse2_loss'] / idx,
'eval_fuse1_loss': val_loss['eval_fuse1_loss'] / idx,
})
trainer.vis.plot_many({
'eval_mask_acc': val_acc['mask_acc'] / idx,
'eval_mask_neg_acc': val_acc['mask_neg_acc'] / idx,
'eval_mask_pos_acc': val_acc['mask_pos_acc'] / idx,
})
# ----------------- save model ---------------- #
if cfg.save_pos_acc < (val_acc['mask_pos_acc'] / idx) and cfg.save_acc < (
val_acc['mask_acc'] / idx):
cfg.save_pos_acc = (val_acc['mask_pos_acc'] / idx)
cfg.save_acc = (val_acc['mask_acc'] / idx)
trainer.saver.save(model, tag='%s_epoch(%d)_acc(%0.5f/%0.5f)' % (
cfg.name, epoch, cfg.save_pos_acc, cfg.save_acc))
trainer.vis.log('Save Model %s_epoch(%d)_acc(%0.5f/%0.5f)' % (
cfg.name, epoch, cfg.save_pos_acc, cfg.save_acc), 'train info')
bar.set_description('Epoch %d --- Training --- :' % epoch)
model.train()
if epoch != 0:
trainer.saver.save(model, tag='%s_epoch(%d)' % (
cfg.name, epoch))
trainer.vis.log('Save Model -%s_epoch(%d)' % (
cfg.name, epoch), 'train info')
except KeyboardInterrupt:
trainer.saver.save(model, tag='Auto_Save_Model')
print('\n Catch KeyboardInterrupt, Auto Save final model : %s' % trainer.saver.show_save_pth_name)
trainer.vis.log('Catch KeyboardInterrupt, Auto Save final model : %s' % trainer.saver.show_save_pth_name,
'train info')
trainer.vis.log('Training End!!')
try:
sys.exit(0)
except SystemExit:
os._exit(0)
if __name__ == '__main__':
main()