python size(0)_对x.view(x.size(0), -1)的一些理解

一般地，在CNN等网络中，都是通过卷积过滤器对目标进行计算，然而这些计算都是建立在高维数据。

最后，项目需要对数据进行分类或者识别，就需要全连接层Linear，这时候就需要将高维数据平铺变为低位数据。

Lenet 例子

class Lenet(nn.Module):

def __init__(self):

super(Lenet, self).__init__()

‘’‘卷积、池化处理’‘’

self.conv = nn.Sequential(

nn.Conv2d(1, 6, 3, stride=1, padding=1),

nn.MaxPool2d(2, 2),

nn.Conv2d(6, 16, 5, stride=1, padding=0),

nn.MaxPool2d(2, 2)

)

‘’‘全连接层’‘’

self.fc = nn.Sequential(

nn.Linear(400, 120),

nn.Linear(120, 84),

nn.Linear(84, 10)

)

def forward(self, x):

out = self.conv(x) #卷积网络(高维数据)

out = out.view(out.size(0), -1)#高维数据 ‘压’成 低维数据

out = self.fc(out)#全连接层(低维数据)

return out

lenet = Lenet()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

例子

import torch as t

a = t.arange(24)

a = a.reshape(4, 3, 2)

x = a.view(a.size(0), -1)

y = a.view(a.size(1), -1)

‘’‘输出对比’‘’

a = tensor([[[ 0, 1],

[ 2, 3],

[ 4, 5]],

[[ 6, 7],

[ 8, 9],

[10, 11]],

[[12, 13],

[14, 15],

[16, 17]],

[[18, 19],

[20, 21],

[22, 23]]])

a.size is torch.Size([4, 3, 2])

x = tensor([[ 0, 1, 2, 3, 4, 5],

[ 6, 7, 8, 9, 10, 11],

[12, 13, 14, 15, 16, 17],

[18, 19, 20, 21, 22, 23]])

x.size is torch.Size([4, 6])

y = tensor([[ 0, 1, 2, 3, 4, 5, 6, 7],

[ 8, 9, 10, 11, 12, 13, 14, 15],

[16, 17, 18, 19, 20, 21, 22, 23]])

y.size is torch.Size([3, 8])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

ps：

x = a.view(a.size(0), -1)

等价于

x = x.view(x.size()[0], -1)1

2

3