-
Notifications
You must be signed in to change notification settings - Fork 1
/
Copy pathmain.py
166 lines (134 loc) · 6.05 KB
/
main.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
# -*- coding: utf-8 -*-
import torch
import torch.nn as nn
import torch.optim as optim
from torch.optim import lr_scheduler
from torch.utils.data import DataLoader
from models.fcn import VGGNet,FCN32s,FCN16s,FCN8s,FCNs
from dataloader.voc import SBDClassSeg
from dataloader.person import Person
from dataloader.own_data import Own_data
from train import train_epoch
from val import val_epoch,eval_own_images
from misc.utils import Cross_Entropy2D,cul_acc
from misc.transform import Trans
import numpy as np
import time
import os
import sys
import argparse
def main():
parser=argparse.ArgumentParser()
parser.add_argument('--id',type=str,default=None,
help='An id to distinguish the saved model')
parser.add_argument('--use_cuda',type=int,default=True)
parser.add_argument('--start_from',type=str,default=None)
# parser.add_argument('--data_type',type=str,default='person')
parser.add_argument('--data_type',type=str,default='own_images')
# parser.add_argument('--data_root',type=str,default='/mnt/disk1/han/dataset/')
# parser.add_argument('--data_root',type=str,default='/mnt/disk1/lihao/person_br/datasets/icome_task2_data')
parser.add_argument('--data_root',type=str,default='/mnt/disk1/lihao/person_br/datasets/own_images/')
parser.add_argument('--optimizer',type=str,default='rmsprop',
help='Choose a optimizer')
parser.add_argument('--max_epochs',type=int,default=40)
parser.add_argument('--max_val_iterations',type=int,default=100,
help='When you don\'t want to test the full val datasets,this help a lot')
parser.add_argument('--batch_size',type=int,default=16)
parser.add_argument('--n_class',type=int,default=2)
parser.add_argument('--lr',type=float,default=5e-4)
parser.add_argument('--momentum',type=float,default=0)
parser.add_argument('--w_decay',type=float,default=1e-5)
parser.add_argument('--update_lr_rate',type=float,default=0.8)
parser.add_argument('--update_lr_step',type=float,default=5)
parser.add_argument('--checkpoint_every',type=int,default=10)
parser.add_argument('--checkpoint_save_path',type=str,default='/mnt/disk1/lihao/person_br/save/')
parser.add_argument('--image_save_path',type=str,default='/mnt/disk1/lihao/person_br/save/imgs4/',
help='Where save the images')
args=parser.parse_args()
kwargs={"num_workers":4,"pin_memory":True} if args.use_cuda else {}
if args.data_type=='voc':
train_loader=DataLoader(SBDClassSeg(args.data_root,
split='train',transform=True),batch_size=args.batch_size,
shuffle=True,**kwargs)
val_loader=DataLoader(SBDClassSeg(args.data_root,
split='val',transform=True),batch_size=args.batch_size,
shuffle=False,**kwargs)
elif args.data_type=='person':
mean_bgr=np.array([128.0523,134.4394,141.8439])
transform=Trans(512,384,mean_bgr)
train_loader=DataLoader(Person(args.data_root,
split='train',transform=transform),batch_size=args.batch_size,
shuffle=True,**kwargs)
val_loader=DataLoader(Person(args.data_root,
split='val',transform=transform),batch_size=args.batch_size,
shuffle=True,**kwargs)
elif args.data_type=='own_images':
mean_bgr=np.array([128.0523,134.4394,141.8439])
transform=Trans(512,384,mean_bgr)
data_loader=DataLoader(Own_data(args.data_root,
transform=transform),batch_size=args.batch_size,
shuffle=False,**kwargs)
else:
raise RuntimeError('No this dataset')
vgg_model=VGGNet(requires_grad=True,remove_fc=True)
model=FCNs(pretrained_net=vgg_model,n_class=args.n_class)
if args.use_cuda:
vgg_model=vgg_model.cuda()
model=model.cuda()
model=nn.DataParallel(model)
criterion=Cross_Entropy2D()
if args.optimizer=='rmsprop':
optimizer=optim.RMSprop(model.parameters(),lr=args.lr,
momentum=args.momentum,weight_decay=args.w_decay)
elif args.optimizer=='adam':
optimizer=optim.Adam(model.parameters(),lr=args.lr,
weight_decay=args.w_decay)
else:
print('Please use adam or rmsprop as your optimizer')
raise RuntimeError('Wrong optimizer')
scheduler=lr_scheduler.StepLR(optimizer,step_size=args.update_lr_step,gamma=args.update_lr_rate)
infos={}
infos['iteration']=0
infos['epoch']=0
infos['train_loss']=[]
infos['val_loss']=[]
infos['mean_Pixel']=[]
infos['meanIU']=[]
if args.start_from is not None and os.path.isfile(args.start_from):
D=torch.load(args.start_from)
if '_best_' in args.start_from:
infos=D['infos']
model.load_state_dict(D['model_state_dict'])
else:
infos=D['infos']
optimizer.load_state_dict(D['optimizer_state_dict'])
scheduler.load_state_dict(D['scheduler_state_dict'])
model.load_state_dict(D['model_state_dict'])
epoch=infos['epoch']
if args.data_type=='own_images':
eval_own_images(model,data_loader,args)
return
best_acc=-1
for i in range(epoch,args.max_epochs):
#Train epoch
train_epoch(model,optimizer,criterion,train_loader,infos,args)
#Val epoch
val_epoch(model,criterion,val_loader,infos,args)
#Save infos and model_dict
scheduler.step()
acc=cul_acc(infos['mean_Pixel'][-1][1],infos['meanIU'][-1][1])
if acc>best_acc:
torch.save({
'infos':infos,
'model_state_dict':model.state_dict(),
},os.path.join(args.checkpoint_save_path,'model_best_'+args.id+'.pkl'))
best_acc=acc
torch.save({
'infos':infos,
'model_state_dict':model.state_dict(),
'optimizer_state_dict':optimizer.state_dict(),
'scheduler_state_dict':scheduler.state_dict(),
},os.path.join(args.checkpoint_save_path,'model_'+args.id+'.pkl'))
infos['epoch']+=1
if __name__=='__main__':
main()