-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathmain.py
193 lines (162 loc) · 8.57 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
186
187
188
189
190
191
192
193
from __future__ import print_function
import torch
import torch.nn as nn
import torch.nn.parallel
import torch.optim
import torch.utils.data
import utils
import load_materials
from models.Model import resnet18_EST
from tensorboardX import SummaryWriter
import pytorch_warmup as warmup
import random
from options.base_options import BaseOptions
from torch.backends import cudnn
import os
def train(train_loader, model, criterion, optimizer, epoch, opt, writer):
running_loss, count, correct_count, running_cls_loss, running_per_loss, correct_per_count = 0., 0, 0., 0., 0., 0.
model.train()
for i, data in enumerate(train_loader):
target_first = data['label']
input_var = torch.autograd.Variable(data['data_shuffle'])
order_label = data['per_label']
target = target_first.cuda(non_blocking=True)
order_label = order_label.cuda(non_blocking=True)
target_var = torch.autograd.Variable(target)
order_label = torch.autograd.Variable(order_label)
pred_score, per_score = model(input_var, per=data['per_shuffle'])
# compute gradient and do Adam step
loss_cls = criterion(pred_score, target_var)
loss_per = criterion(per_score, order_label)
loss = loss_cls + loss_per * opt.lamb
optimizer.zero_grad()
loss.backward()
optimizer.step()
# store loss
running_loss += loss.item()
running_cls_loss += loss_cls.item()
running_per_loss += loss_per.item()
correct_count += (torch.max(pred_score, dim=1)[1] == target_var).sum()
correct_per_count += (torch.max(per_score, dim=1)[1] == order_label).sum()
count += input_var.size(0)
if i % opt.print_freq == 0:
print(
'Epoch: [{0}][{1}/{2}]\t Loss {loss:.4f}\t Cls_Acc{acc:.4f}\t Per_Acc{per_acc:.4f}\t Loss cls {loss_cls:.4f}\t Loss per {loss_per:.4f}'
.format(epoch, i, len(train_loader), loss=running_loss / count, acc=int(correct_count) / count,
per_acc=int(correct_per_count) / count, loss_cls=running_cls_loss / count,
loss_per=running_per_loss / count))
print(
' Train_Acc {train_Video:.4f}\t Train_Loss {Train_Loss:.4f}\t Per_Acc{per_acc:.4f}\t Loss cls {loss_cls:.4f}\t Loss per {loss_per:.4f}'.
format(train_Video=int(correct_count) / count, Train_Loss=running_loss / count,
per_acc=int(correct_per_count) / count, loss_cls=running_cls_loss / count,
loss_per=running_per_loss / count))
writer.add_scalar('final_loss', running_loss / count, epoch)
writer.add_scalar('final_cls_loss', running_cls_loss / count, epoch)
writer.add_scalar('final_cls_acc', int(correct_count) / count, epoch)
writer.add_scalar('final_per_loss', running_per_loss / count, epoch)
writer.add_scalar('final_per_acc', int(correct_per_count) / count, epoch)
def validate(val_loader, model,args):
model.eval()
test_correct_count, test_count, test_correct_per_count, test_per_acc = 0, 0, 0, 0.
with torch.no_grad():
for i, data in enumerate(val_loader):
########################test shuffle
input_var = torch.autograd.Variable(data['data_shuffle'])
order_label = data['per_label']
order_label = order_label.cuda(non_blocking=True)
order_label = torch.autograd.Variable(order_label)
# compute output
_, per_score = model(input_var, per=data['per_shuffle'])
####################################################
#################### test cls
input_var = torch.autograd.Variable(data['data_normal'])
target_first = data['label']
# compute output
target = target_first.cuda(non_blocking=True)
target_var = torch.autograd.Variable(target)
pred_score, _ = model(input_var, per=data['per_normal'])
#####################################################
#if torch.max(pred_score, dim=1)[1] != target_var:
# print(data['path'], ' ', utils.cate2label(args.dataset_name)[torch.max(pred_score, dim=1)[1].item()])
test_correct_count += (torch.max(pred_score, dim=1)[1] == target_var).sum()
test_correct_per_count += (torch.max(per_score, dim=1)[1] == order_label).sum()
test_count += input_var.size(0)
if args.draw_weight:
video_path_list = data['path'][0].split('/')
video_path = video_path_list[-2]+'/'+video_path_list[-1]
heat_map_path = os.path.join(args.heat_map_path,args.name,video_path)
trans_path = os.path.join(heat_map_path,'trans')
cos_path = os.path.join(heat_map_path,'cos')
utils.mkdirs(trans_path)
utils.mkdirs(cos_path)
for t in range(len(weight_list)):
heat = os.path.join(trans_path,str(t)+'.npy')
utils.draw_weight(weight_list[t].squeeze().detach().cpu().numpy(),heat)
for t in range(len(cos_weight)):
heat = os.path.join(cos_path,str(t)+'.npy')
utils.draw_weight(cos_weight[t].detach().cpu().numpy(),heat)
test_acc = int(test_correct_count) / test_count
test_per_acc = int(test_correct_per_count) / test_count
print(' Test_Acc: {test_Video:.4f} '.format(test_Video=test_acc))
print(' Test_per_Acc: {test_per_Video:.4f} '.format(test_per_Video=test_per_acc))
return test_acc, test_per_acc
def main(opt):
train_loader, val_loader = load_materials.LoadDataset(opt)
model = resnet18_EST(clips=opt.snippets, img_num_per_clip=opt.per_snippets, d_model=opt.d_model, nhead=opt.nhead,
encoder_nums=opt.encoder_nums, decoder_nums=opt.decoder_nums, use_norm=opt.use_norm,
per_classes=opt.permutation_classes,draw_weight=opt.draw_weight)
if opt.isTrain and not opt.continue_train:
model = load_materials.LoadParameter(model, opt.parameterDir)
print('train !')
elif opt.continue_train:
model = torch.nn.DataParallel(model).cuda()
model.load_state_dict(torch.load(opt.pre_train_model_path)['state_dict'])
print('load eval model !')
else:
print('load eval model !')
model = torch.nn.DataParallel(model).cuda()
model.load_state_dict(torch.load(opt.eval_model_path)['state_dict'])
criterion = nn.CrossEntropyLoss().cuda()
cudnn.benchmark = True
if not opt.isTrain:
validate(val_loader, model,opt)
return
per_branch_params = list(map(id, model.module.per_branch.parameters()))
base_params = filter(lambda p: id(p) not in per_branch_params and p.requires_grad, model.parameters())
optimizer = torch.optim.Adam([
{'params': base_params},
{'params': model.module.per_branch.parameters(), 'lr': opt.lr}
], lr=opt.lr, betas=(0.9, 0.999), weight_decay=opt.weight_decay)
lr_schduler = torch.optim.lr_scheduler.CosineAnnealingLR(optimizer, T_max=opt.epochs_count)
warmup_scheduler = warmup.LinearWarmup(optimizer, warmup_period=opt.warm_up)
warmup_scheduler.last_step = -1
best_prec1 = 0.
for epoch in range(opt.epoch, opt.epochs_count):
lr_schduler.step(epoch)
warmup_scheduler.dampen()
train(train_loader, model, criterion, optimizer, epoch, opt, writer)
prec1, per_acc = validate(val_loader, model,opt)
writer.add_scalar('final_test_acc', prec1, epoch)
writer.add_scalar('final_test_per_acc', per_acc, epoch)
is_best = prec1 > best_prec1
if is_best:
print('better model!')
best_prec1 = max(prec1, best_prec1)
utils.save_checkpoint({
'epoch': epoch + 1,
'state_dict': model.state_dict(),
'prec1': prec1,
}, opt)
else:
print('Model too bad & not save')
if __name__ == '__main__':
opt = BaseOptions().parse()
cudnn.benchmark = False # if benchmark=True, deterministic will be False
cudnn.deterministic = True
torch.manual_seed(opt.seed) # 为CPU设置随机种子
torch.cuda.manual_seed(opt.seed) # 为当前GPU设置随机种子
torch.cuda.manual_seed_all(opt.seed) # 为所有GPU设置随机种子
random.seed(opt.seed)
writer = SummaryWriter(comment=opt.name)
main(opt)
writer.close()