forked from kaijieshi7/Dynamic-convolution-Pytorch
-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.py
130 lines (114 loc) · 5.84 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
import torch.optim as optim
import torch.nn as nn
import torch.nn.functional as F
import torch
import time
import argparse
import torchvision.models.vgg
import torchvision.transforms as transforms
from dynamic_models.dy_vgg import vgg11 as dy_vgg11
from dynamic_models.raw_vgg import vgg11 as raw_vgg11
from dynamic_models.dy_resnet import resnet18 as dy_resnet18
from torchvision.models.resnet import resnet18 as raw_resnet18
parser = argparse.ArgumentParser(description='dynamic convolution')
parser.add_argument('--dataset', type=str, default='cifar10', help='training dataset')
parser.add_argument('--batch-size', type=int, default=128)
parser.add_argument('--test-batch-size', type=int, default=20)
parser.add_argument('--epochs', type=int, default=160)
parser.add_argument('--lr', type=float, default=0.1, )
parser.add_argument('--momentum', type=float, default=0.9)
parser.add_argument('--weight-decay', type=float, default=1e-4)
parser.add_argument('--net-name', default='vgg16_bn')
args = parser.parse_args()
args.device = 'cuda:0' if torch.cuda.is_available() else 'cpu'
if args.dataset == 'cifar10':
numclasses=10
trainset = torchvision.datasets.CIFAR10(root='./data', train=True, download=True,
transform=transforms.Compose([
transforms.Pad(4),
transforms.RandomCrop(32),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465),(0.2023, 0.1994, 0.2010))
]))
trainloader = torch.utils.data.DataLoader(trainset, batch_size=args.batch_size, shuffle=True, num_workers=0)
testset = torchvision.datasets.CIFAR10(root='./data', train=False, download=True,
transform=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))
]))
testloader = torch.utils.data.DataLoader(testset, batch_size=args.batch_size, shuffle=False, num_workers=0)
elif args.dataset=='cifar100':
numclasses=100
trainset = torchvision.datasets.CIFAR100(root='./data', train=True, download=True,
transform=transforms.Compose([
transforms.Pad(4),
transforms.RandomCrop(32),
transforms.RandomHorizontalFlip(),
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))
]))
trainloader = torch.utils.data.DataLoader(trainset, batch_size=args.batch_size, shuffle=True, num_workers=0)
testset = torchvision.datasets.CIFAR100(root='./data', train=False, download=True,
transform=transforms.Compose([
transforms.ToTensor(),
transforms.Normalize((0.4914, 0.4822, 0.4465), (0.2023, 0.1994, 0.2010))
]))
testloader = torch.utils.data.DataLoader(testset, batch_size=args.batch_size, shuffle=False, num_workers=0)
if args.net_name=='dy_resnet18':
model = dy_resnet18(num_classes=numclasses)
elif args.net_name=='raw_resnet18':
model = raw_resnet18(num_classes=numclasses)
elif args.net_name=='raw_vgg11':
model = raw_vgg11(num_classes=numclasses)
elif args.net_name=='dy_vgg11':
model = dy_vgg11(num_classes=numclasses)
model.to(args.device)
optimizer = optim.SGD(model.parameters(), lr=args.lr, momentum=args.momentum, weight_decay=args.weight_decay)
print(str(args))
def adjust_lr(optimizer, epoch):
if epoch in [args.epochs*0.5, args.epochs*0.75, args.epochs*0.85]:
for p in optimizer.param_groups:
p['lr'] *= 0.1
lr = p['lr']
print('Change lr:'+str(lr))
def train(epoch):
model.train()
avg_loss = 0.
train_acc = 0.
adjust_lr(optimizer, epoch)
for batch_idx, (data, target) in enumerate(trainloader):
data, target = data.to(args.device), target.to(args.device)
optimizer.zero_grad()
output = model(data)
loss = F.cross_entropy(output, target)
avg_loss += loss.item()
pred = output.data.max(1, keepdim=True)[1]
train_acc += pred.eq(target.data.view_as(pred)).cpu().sum()
loss.backward()
optimizer.step()
print('Train Epoch: {}, loss{:.6f}, acc{}'.format(epoch, loss.item(), train_acc/len(trainloader.dataset)), end='')
if args.net_name.startswith('dy'):
model.update_temperature()
def val(epoch):
model.eval()
test_loss = 0.
correct=0.
with torch.no_grad():
for data, label in testloader:
data, label = data.to(args.device), label.to(args.device)
output = model(data)
test_loss += F.cross_entropy(output, label, size_average=False).item()
pred = output.data.max(1, keepdim=True)[1]
correct += pred.eq(label.data.view_as(pred)).cpu().sum()
test_loss/=len(testloader.dataset)
correct = int(correct)
print('Test set:average loss: {:.4f}, accuracy{}'.format(test_loss, 100.*correct/len(testloader.dataset)))
return correct/len(testloader.dataset)
best_val_acc=0.
for i in range(args.epochs):
train(i+1)
temp_acc = val(i+1)
if temp_acc>best_val_acc:
best_val_acc = temp_acc
print('Best acc{}'.format(best_val_acc))