-
Notifications
You must be signed in to change notification settings - Fork 63
/
Copy pathmain.py
185 lines (177 loc) · 7.66 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
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
# coding:utf-8
from __future__ import print_function
import torch
import torch.nn as nn
import torch.nn.functional as F
from torch.autograd import Variable
from torchvision import transforms
from torch.utils.data import Dataset, DataLoader
import datetime
#------------------------
from utils import *
import cfgs_hw as cfgs
#------------------------
def display_cfgs(models):
print('global_cfgs')
cfgs.showcfgs(cfgs.global_cfgs)
print('dataset_cfgs')
cfgs.showcfgs(cfgs.dataset_cfgs)
print('net_cfgs')
cfgs.showcfgs(cfgs.net_cfgs)
print('optimizer_cfgs')
cfgs.showcfgs(cfgs.optimizer_cfgs)
print('saving_cfgs')
cfgs.showcfgs(cfgs.saving_cfgs)
for model in models:
print(model)
def flatten_label(target):
label_flatten = []
label_length = []
for i in range(0, target.size()[0]):
cur_label = target[i].tolist()
label_flatten += cur_label[:cur_label.index(0)+1]
label_length.append(cur_label.index(0)+1)
label_flatten = torch.LongTensor(label_flatten)
label_length = torch.IntTensor(label_length)
return (label_flatten, label_length)
def Train_or_Eval(models, state = 'Train'):
for model in models:
if state == 'Train':
model.train()
else:
model.eval()
def Zero_Grad(models):
for model in models:
model.zero_grad()
def Updata_Parameters(optimizers, frozen):
for i in range(0, len(optimizers)):
if i not in frozen:
optimizers[i].step()
#---------------------dataset
def load_dataset():
train_data_set = cfgs.dataset_cfgs['dataset_train'](**cfgs.dataset_cfgs['dataset_train_args'])
train_loader = DataLoader(train_data_set, **cfgs.dataset_cfgs['dataloader_train'])
test_data_set = cfgs.dataset_cfgs['dataset_test'](**cfgs.dataset_cfgs['dataset_test_args'])
test_loader = DataLoader(test_data_set, **cfgs.dataset_cfgs['dataloader_test'])
# pdb.set_trace()
return (train_loader, test_loader)
#---------------------network
def load_network():
model_fe = cfgs.net_cfgs['FE'](**cfgs.net_cfgs['FE_args'])
cfgs.net_cfgs['CAM_args']['scales'] = model_fe.Iwantshapes()
model_cam = cfgs.net_cfgs['CAM'](**cfgs.net_cfgs['CAM_args'])
model_dtd = cfgs.net_cfgs['DTD'](**cfgs.net_cfgs['DTD_args'])
if cfgs.net_cfgs['init_state_dict_fe'] != None:
model_fe.load_state_dict(torch.load(cfgs.net_cfgs['init_state_dict_fe']))
if cfgs.net_cfgs['init_state_dict_cam'] != None:
model_cam.load_state_dict(torch.load(cfgs.net_cfgs['init_state_dict_cam']))
if cfgs.net_cfgs['init_state_dict_dtd'] != None:
model_dtd.load_state_dict(torch.load(cfgs.net_cfgs['init_state_dict_dtd']))
model_fe.cuda()
model_cam.cuda()
model_dtd.cuda()
return (model_fe, model_cam, model_dtd)
#----------------------optimizer
def generate_optimizer(models):
out = []
scheduler = []
for i in range(0, len(models)):
out.append(cfgs.optimizer_cfgs['optimizer_{}'.format(i)](
models[i].parameters(),
**cfgs.optimizer_cfgs['optimizer_{}_args'.format(i)]))
scheduler.append(cfgs.optimizer_cfgs['optimizer_{}_scheduler'.format(i)](
out[i],
**cfgs.optimizer_cfgs['optimizer_{}_scheduler_args'.format(i)]))
return tuple(out), tuple(scheduler)
#---------------------testing stage
def test(test_loader, model, tools):
Train_or_Eval(model, 'Eval')
for sample_batched in test_loader:
data = sample_batched['image']
label = sample_batched['label']
target = tools[0].encode(label)
data = data.cuda()
target = target
label_flatten, length = tools[1](target)
target, label_flatten = target.cuda(), label_flatten.cuda()
features= model[0](data)
A = model[1](features)
output, out_length = model[2](features[-1], A, target, length, True)
tools[2].add_iter(output, out_length, length, label)
tools[2].show()
Train_or_Eval(model, 'Train')
#---------------------------------------------------------
#--------------------------Begin--------------------------
#---------------------------------------------------------
if __name__ == '__main__':
# prepare nets, optimizers and data
model = load_network()
display_cfgs(model)
optimizers, optimizer_schedulers = generate_optimizer(model)
criterion_CE = nn.CrossEntropyLoss().cuda()
train_loader, test_loader = load_dataset()
print('preparing done')
# --------------------------------
# prepare tools
train_acc_counter = Attention_AR_counter('train accuracy: ', cfgs.dataset_cfgs['dict_dir'], cfgs.dataset_cfgs['case_sensitive'])
test_acc_counter = Attention_AR_counter('\ntest accuracy: ', cfgs.dataset_cfgs['dict_dir'], cfgs.dataset_cfgs['case_sensitive'])
loss_counter = Loss_counter(cfgs.global_cfgs['show_interval'])
encdec = cha_encdec(cfgs.dataset_cfgs['dict_dir'], cfgs.dataset_cfgs['case_sensitive'])
#---------------------------------
if cfgs.global_cfgs['state'] == 'Test':
test((test_loader),
model,
[encdec,
flatten_label,
test_acc_counter])
exit()
# --------------------------------
total_iters = len(train_loader)
for nEpoch in range(0, cfgs.global_cfgs['epoch']):
for batch_idx, sample_batched in enumerate(train_loader):
# data prepare
data = sample_batched['image']
label = sample_batched['label']
target = encdec.encode(label)
Train_or_Eval(model, 'Train')
data = data.cuda()
label_flatten, length = flatten_label(target)
target, label_flatten = target.cuda(), label_flatten.cuda()
# net forward
features = model[0](data)
A = model[1](features)
output, attention_maps = model[2](features[-1], A, target, length)
# computing accuracy and loss
train_acc_counter.add_iter(output, length.long(), length, label)
loss = criterion_CE(output, label_flatten)
loss_counter.add_iter(loss)
# update network
Zero_Grad(model)
loss.backward()
nn.utils.clip_grad_norm_(model[0].parameters(), 20, 2)
nn.utils.clip_grad_norm_(model[1].parameters(), 20, 2)
nn.utils.clip_grad_norm_(model[2].parameters(), 20, 2)
Updata_Parameters(optimizers, frozen = [])
# visualization and saving
if batch_idx % cfgs.global_cfgs['show_interval'] == 0 and batch_idx != 0:
print(datetime.datetime.now().strftime('%H:%M:%S'))
print('Epoch: {}, Iter: {}/{}, Loss dan: {}'.format(
nEpoch,
batch_idx,
total_iters,
loss_counter.get_loss()))
train_acc_counter.show()
if batch_idx % cfgs.global_cfgs['test_interval'] == 0 and batch_idx != 0:
test((test_loader),
model,
[encdec,
flatten_label,
test_acc_counter])
if nEpoch % cfgs.saving_cfgs['saving_epoch_interval'] == 0 and \
batch_idx % cfgs.saving_cfgs['saving_iter_interval'] == 0 and \
batch_idx != 0:
for i in range(0, len(model)):
torch.save(model[i].state_dict(),
cfgs.saving_cfgs['saving_path'] + 'E{}_I{}-{}_M{}.pth'.format(
nEpoch, batch_idx, total_iters, i))
Updata_Parameters(optimizer_schedulers, frozen = [])