-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
157 lines (145 loc) · 6.28 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
# -*- coding: utf-8 -*-
"""
@File ：train.py
@Time ： 2023/1/10 0010 14:41
@Auth ： GiserLee
@E-mail：554758017@qq.com
@IDE ：PyCharm
"""
import torch
import torch.nn as nn
from tqdm import tqdm
import numpy as np
from corecode.dataset.CDdataSet import CDdataSet
from corecode.dataset.CDdataLoader import CDDataLoader
from corecode.models import UNet_CD,UNet_ASPP,SiamUnet_conc
from corecode.loss.loss_func import HybirdLoss
from corecode.metrics.confunsionmetrics import CDMetric
from torch.optim import lr_scheduler
# 加载模型,确定是否使用初始化方法
def load_model(model,inputimage_CH,classNum,device,is_init=True):
'''
inputimage_CH:
if is siam type model,inputimage_CH=input_channels (3)
else inputimage_CH=input_channels*2 (6)
'''
model = model(input_nbr=inputimage_CH, label_nbr=classNum)
if is_init:
for m in model.modules():
if isinstance(m, nn.Conv2d):
nn.init.kaiming_uniform_(m.weight, mode="fan_in", nonlinearity="relu")
if m.bias is not None:
nn.init.constant_(m.bias, 0)
elif isinstance(m, nn.BatchNorm2d):
nn.init.constant_(m.weight, 1)
nn.init.constant_(m.bias, 0)
model.to(device)
return model
#保存模型
def save_model(exp_dir, model, optimizer):
"""
Model Saving Function
Args:
exp_dir (Path) : model saving path
model : torch model
optimizer : torch optimizer
"""
torch.save(
{
'model_state_dict': model.state_dict(),
'optimize_state_dict': optimizer.state_dict(),
},
exp_dir
)
def single_train(model,data_loader,optimizer,loss_fun,device):
"""
Train Function for a single Epoch
"""
model.train()
len_loader = len(data_loader)
total_loss = 0.
criterion_list = []
train_metrics = CDMetric(numClass=2)
#for i,(A, B, Target) in enumerate(data_loader):
pbar = tqdm(data_loader)
for i,(A, B, Target) in enumerate(pbar):
pbar.set_description(f'Train Iter good luck:')
A, B, Target = A.to(device,dtype=torch.float), B.to(device,dtype=torch.float), Target.to(device,dtype=torch.long)
optimizer.zero_grad()
output = model(A,B)
Target = torch.squeeze(Target,dim=1) # b 1 h w => b h w
loss = loss_fun(output, Target)
#loss = FocalLoss()(output, Target)
loss.backward()
optimizer.step()
total_loss += loss.item()
"""精度评定"""
train_metrics.reset()
train_metrics.addBatchNew(imgPredict=output,imgLabel=Target)
criterion_list.append(train_metrics.pre_reca_fscor())
if i % 2 == 0:
iter_precision,iter_recall,iter_f1 =np.sum(criterion_list,axis=0)
pbar.set_postfix({'Loss': total_loss / (i+1),'Pre': iter_precision / (i+1)
,'Recall': iter_recall / (i+1),'F1': iter_f1 / (i+1)})
total_loss /= len_loader
# epoc_precision,epoc_recall,epoc_f1=np.sum(criterion_list,axis=0)
# print(f'train result-Iter :{len_loader}/{len_loader} Loss :{total_loss:.3f} Pre :{epoc_precision/len_loader:.3f} F1 :{epoc_f1/len_loader:.3f} R :{epoc_recall/len_loader:.3f}')
return total_loss
def single_val(model,data_loader,loss_fun,device):
"""
Validation Function for a single Epoch
"""
model.eval()
len_loader = len(data_loader)
total_loss = 0.
criterion_list = []
val_metrics = CDMetric(numClass=2)
with torch.no_grad():
pbar = tqdm(data_loader)
for i, (A, B, Target) in enumerate(pbar):
pbar.set_description('star val please waite')
A, B, Target = A.to(device,dtype=torch.float), B.to(device,dtype=torch.float), Target.to(device,dtype=torch.long)
output = model(A, B)
Target = torch.squeeze(Target, dim=1)
loss = loss_fun(output, Target)
total_loss += loss.item()
val_metrics.reset()
val_metrics.addBatchNew(imgPredict=output, imgLabel=Target)
criterion_list.append(val_metrics.pre_reca_fscor())
if (i+1) == len_loader:
epoc_precision, epoc_recall, epoc_f1 = np.sum(criterion_list, axis=0)
pbar.set_postfix({'Loss':total_loss/len_loader,'Pre':epoc_precision / len_loader,'F1':epoc_f1 / len_loader,'Rrcall':epoc_recall / len_loader})
#print(f'val result-Iter :{len_loader}/{len_loader} Loss :{total_loss:.3f} Pre :{epoc_precision / len_loader:.3f} F1 :{epoc_f1 / len_loader:.3f} Rrcall :{epoc_recall / len_loader:.3f}')
total_loss /= len_loader
lossval = [total_loss,epoc_precision/len_loader,epoc_f1/len_loader,epoc_recall/len_loader]
return lossval
if __name__ == '__main__':
DEVICE = 'cuda:0' if torch.cuda.is_available() else 'cpu'
data_path = r'F:\dataset\LEVIRCD\CDLabDataset'
model_save_path = r'F:\changedetection_code\Myconde\corecode\save\model\cdUnet.pth'
batch_size = 2
input_channels = 3
class_num = 2
lr_rate = 5e-3
num_epochs = 100
train_dataset = CDdataSet(data_path=data_path,modeOfdir='train',test_mode=False)
val_dataset = CDdataSet(data_path=data_path,modeOfdir='val',test_mode=True)
train_dataLoader = CDDataLoader(dataset=train_dataset,batch_size=batch_size)
val_dataLoader = CDDataLoader(dataset=val_dataset,batch_size=batch_size)
model = load_model(model=UNet_CD,inputimage_CH=input_channels*2,classNum=class_num,device=DEVICE)
# criterion = torch.nn.CrossEntropyLoss().to(DEVICE)
criterion = HybirdLoss(class_num=class_num)
optimizer = torch.optim.Adam(model.parameters(),lr=lr_rate)
scheduler = lr_scheduler.StepLR(optimizer, step_size=30, gamma=0.1)
best_loss_precision = 0
for i in range(num_epochs):
single_train(model=model,data_loader=train_dataLoader,optimizer=optimizer
,loss_fun=criterion,device=DEVICE)
scheduler.step()
valloss = single_val(model=model,data_loader=val_dataLoader,loss_fun=criterion,device=DEVICE)
precision = valloss[1]
"""model save """
if best_loss_precision < precision:
best_loss_precision = precision
save_model(exp_dir=model_save_path,model=model,optimizer=optimizer)
print("valid precision is improved. the model is saved.")