-
Notifications
You must be signed in to change notification settings - Fork 13
/
Copy pathtrain_LF_2.py
216 lines (204 loc) · 10.2 KB
/
train_LF_2.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
204
205
206
207
208
209
210
211
212
213
214
215
216
# coding:utf-8
from __future__ import print_function
import torch
import torch.nn as nn
from torch.utils.data import Dataset, DataLoader
import datetime
from utils import *
import cfgs.cfgs_LF_2 as cfgs
from collections import OrderedDict
import time
import sys
import os
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(model, state='Train'):
if state == 'Train':
model.train()
else:
model.eval()
def Zero_Grad(model):
model.zero_grad()
def Updata_Parameters(optimizers, frozen):
for i in range(0, len(optimizers)):
if i not in frozen:
optimizers[i].step()
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'])
return train_loader, test_loader
def load_network():
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
model_VL = cfgs.net_cfgs['VisualLAN'](**cfgs.net_cfgs['args'])
model_VL = model_VL.to(device)
model_VL = torch.nn.DataParallel(model_VL)
if cfgs.net_cfgs['init_state_dict'] != None:
fe_state_dict_ori = torch.load(cfgs.net_cfgs['init_state_dict'])
fe_state_dict = OrderedDict()
for k, v in fe_state_dict_ori.items():
if 'module' not in k:
k = 'module.' + k
else:
k = k.replace('features.module.', 'module.features.')
fe_state_dict[k] = v
model_dict_fe = model_VL.state_dict()
state_dict_fe = {k: v for k, v in fe_state_dict.items() if k in model_dict_fe.keys()}
model_dict_fe.update(state_dict_fe)
model_VL.load_state_dict(model_dict_fe)
return model_VL
def generate_optimizer(model):
if cfgs.global_cfgs['step'] != 'LF_2':
out = torch.optim.Adam([{'params': model.parameters(), 'lr': cfgs.optimizer_cfgs['optimizer_0_args']['lr']}])
scheduler = cfgs.optimizer_cfgs['optimizer_0_scheduler'](out, **cfgs.optimizer_cfgs['optimizer_0_scheduler_args'])
return out, scheduler
else:
id_mlm = id(model.module.MLM_VRM.MLM.parameters())
id_pre_mlm = id(model.module.MLM_VRM.Prediction.pp_share.parameters()) + id(model.module.MLM_VRM.Prediction.w_share.parameters())
id_total = id_mlm + id_pre_mlm
out = torch.optim.Adam([{'params': filter(lambda p: id(p) == id_total, model.parameters()), 'lr': cfgs.optimizer_cfgs['optimizer_0_args']['lr']},
{'params': filter(lambda p: id(p) != id_total, model.parameters()),'lr': cfgs.optimizer_cfgs['optimizer_0_args']['lr'] * 0.1}])
scheduler = cfgs.optimizer_cfgs['optimizer_0_scheduler'](out, **cfgs.optimizer_cfgs['optimizer_0_scheduler_args'])
return out, scheduler
def _flatten(sources, lengths):
return torch.cat([t[:l] for t, l in zip(sources, lengths)])
def test(test_loader, model, tools, best_acc):
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()
output, out_length = model(data, target, '', False)
tools[2].add_iter(output, out_length, length, label)
best_acc, change = tools[2].show_test(best_acc)
Train_or_Eval(model, 'Train')
return best_acc, change
if __name__ == '__main__':
model = load_network()
optimizer, optimizer_scheduler = generate_optimizer(model)
criterion_CE = nn.CrossEntropyLoss().cuda()
L1_loss = nn.L1Loss().cuda()
train_loader, test_loader = load_dataset()
# tools prepare
train_acc_counter = Attention_AR_counter('train accuracy: ', cfgs.dataset_cfgs['dict_dir'],
cfgs.dataset_cfgs['case_sensitive'])
train_acc_counter_rem = Attention_AR_counter('train accuracy: ', cfgs.dataset_cfgs['dict_dir'],
cfgs.dataset_cfgs['case_sensitive'])
train_acc_counter_sub = 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'])
encdec = cha_encdec(cfgs.dataset_cfgs['dict_dir'], cfgs.dataset_cfgs['case_sensitive'])
# train
total_iters = len(train_loader)
loss_show = 0
time_cal = 0
ratio_res = 0.5
ratio_sub = 0.5
best_acc = 0
loss_ori_show = 0
loss_mas_show = 0
if not os.path.isdir(cfgs.saving_cfgs['saving_path']):
os.mkdir(cfgs.saving_cfgs['saving_path'])
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'] # original string
label_res = sample_batched['label_res'] # remaining string
label_sub = sample_batched['label_sub'] # occluded character
label_id = sample_batched['label_id'] # character index
target = encdec.encode(label)
target_res = encdec.encode(label_res)
target_sub = encdec.encode(label_sub)
Train_or_Eval(model, 'Train')
data = data.cuda()
label_flatten, length = flatten_label(target)
label_flatten_res, length_res = flatten_label(target_res)
label_flatten_sub, length_sub = flatten_label(target_sub)
target, label_flatten, target_res, target_sub, label_flatten_res = target.cuda(), label_flatten.cuda(), target_res.cuda(), target_sub.cuda(), label_flatten_res.cuda()
label_flatten_sub, label_id = label_flatten_sub.cuda(), label_id.cuda()
# prediction
text_pre, text_rem, text_mas, att_mask_sub = model(data, label_id, cfgs.global_cfgs['step'])
# loss_calculation
if cfgs.global_cfgs['step'] == 'LF_1':
text_pre = _flatten(text_pre, length)
pre_ori, label_ori = train_acc_counter.add_iter(text_pre, length.long(), length, label)
loss_ori = criterion_CE(text_pre, label_flatten)
loss = loss_ori
else:
text_pre = _flatten(text_pre, length)
text_rem = _flatten(text_rem, length_res)
text_mas = _flatten(text_mas, length_sub)
pre_ori, label_ori = train_acc_counter.add_iter(text_pre, length.long(), length, label)
pre_rem, label_rem = train_acc_counter_rem.add_iter(text_rem, length_res.long(), length_res, label_res)
pre_sub, label_sub = train_acc_counter_sub.add_iter(text_mas, length_sub.long(), length_sub, label_sub)
loss_ori = criterion_CE(text_pre, label_flatten)
loss_res = criterion_CE(text_rem, label_flatten_res)
loss_mas = criterion_CE(text_mas, label_flatten_sub)
loss = loss_ori + loss_res * ratio_res + loss_mas * ratio_sub
loss_ori_show += loss_res
loss_mas_show += loss_mas
# loss for display
loss_show += loss
# optimize
Zero_Grad(model)
loss.backward()
nn.utils.clip_grad_norm_(model.parameters(), 20, 2)
optimizer.step()
# display
if batch_idx % cfgs.global_cfgs['show_interval'] == 0 and batch_idx != 0:
loss_show = loss_show / cfgs.global_cfgs['show_interval']
print(datetime.datetime.now().strftime('%H:%M:%S'))
print(
'Epoch: {}, Iter: {}/{}, Loss VisionLAN: {:0.4f}'.format(
nEpoch,
batch_idx,
total_iters,
loss_show))
loss_show = 0
train_acc_counter.show()
if cfgs.global_cfgs['step'] != 'LF_1':
print(
'orignial: {}, mask_character pre/gt: {}/{}, other pre/gts: {}/{}'.format(
label[0],
pre_sub[0],
label_sub[0],
pre_rem[0],
label_rem[0]))
loss_mas_show = loss_mas_show / cfgs.global_cfgs['show_interval']
loss_ori_show = loss_ori_show / cfgs.global_cfgs['show_interval']
print('loss for mas/rem: {}/{}'.format(loss_mas_show,loss_ori_show))
loss_ori_show = 0
loss_mas_show = 0
sys.stdout.flush()
# evaluation during training
if batch_idx % cfgs.global_cfgs['test_interval'] == 0 and batch_idx != 0:
print('Testing during training:')
best_acc, if_save = test((test_loader),
model,
[encdec,
flatten_label,
test_acc_counter], best_acc)
if if_save:
torch.save(model.state_dict(),
cfgs.saving_cfgs['saving_path'] + 'best_acc_M.pth')
# save each epoch
if nEpoch % cfgs.saving_cfgs['saving_epoch_interval'] == 0:
torch.save(model.state_dict(),
cfgs.saving_cfgs['saving_path'] + 'E{}.pth'.format(
nEpoch))
optimizer_scheduler.step()