pytorch(一)一个简单的CNN搭建

冷不防 2023-01-05 03:58 147阅读 0赞

网络组成：卷积、激活、池化、BN、dropout、全连接层等。[各层作用][Link 1]。

需要设定某些网络层的相关参数inchannel,outchannel,kernel\_size、padding、stride等。

若输入数据维度为W\*W   kernelsize： *F×F*     stride：*S*          padding： *P*

可以得出卷积层输出的特征图尺寸计算：N = (*W − F* + 2*P* )/*S*\+1

nn.MaxPool2d()函数中，参数ceil\_mode默认为False表示向下取整，如果等于True，向上取整

池化层输出计算公式：dilation默认为1，如果不设置stride则默认为kernelsize大小。[来源][Link 2]

![2021011220580876.png][]

例子：

假设输入：3\*448\*448

经过5次卷积要获得256\*14\*14 feature map（fp）

最简单方法是卷积层stride与padding都为1，核为3；池化层核为2，stride为2。

import torch.nn as nn
    import torch
    import numpy as np
    import torch.nn.functional as F
    
    class myNet(nn.Module):
        def __init__(self):
            super(myNet, self).__init__()
            self.conv1=nn.Conv2d(3,32,3,1,1)
            self.conv2 = nn.Conv2d(32, 64, 3,1,1)
            self.conv3 = nn.Conv2d(64, 128, 3,1,1)
            self.conv4 = nn.Conv2d(128, 128, 3,1,1)
            self.conv5 = nn.Conv2d(128, 256, 3,1,1)
            self.linear1=nn.Linear(256*14*14,4096)
            self.linear2=nn.Linear(4096,4096)
            self.linear3=nn.Linear(4096,10)
    
        def forward(self,x):
            x = F.max_pool2d(F.relu(self.conv1(x)),(2,2))
            x = F.max_pool2d(F.relu(self.conv2(x)), (2, 2))
            x = F.max_pool2d(F.relu(self.conv3(x)), (2, 2))
            x = F.max_pool2d(F.relu(self.conv4(x)), (2, 2))
            x = F.max_pool2d(F.relu(self.conv5(x)), (2, 2))
            print(x.size())
            x = x.view(x.size()[0], -1)
            x = self.linear1(x)
            print(x.shape)
            x = self.linear2(x)
            print(x.shape)
            x = self.linear3(x)
    
            return x
    
    
    
    inputs=torch.randn(1,3,448,448)
    model=myNet()
    #打印各层网络参数信息
    print(model)
    print(model(inputs).shape)
    
    
    
    """
    输出信息：
    myNet(
      (conv1): Conv2d(3, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (conv2): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (conv3): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (conv4): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (conv5): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))
      (linear1): Linear(in_features=50176, out_features=4096, bias=True)
      (linear2): Linear(in_features=4096, out_features=4096, bias=True)
      (linear3): Linear(in_features=4096, out_features=10, bias=True)
    )
    torch.Size([1, 256, 14, 14])
    torch.Size([1, 4096])
    torch.Size([1, 4096])
    torch.Size([1, 10])
    
    
    """

待续。。。

[Link 1]: https://blog.csdn.net/weixin_43843657/article/details/89138646?utm_medium=distribute.pc_relevant.none-task-blog-BlogCommendFromBaidu-3.control&depth_1-utm_source=distribute.pc_relevant.none-task-blog-BlogCommendFromBaidu-3.control
[Link 2]: https://blog.csdn.net/shanglianlm/article/details/85313924
[2021011220580876.png]: /images/20221119/b301f392551648a3a5528d6ca385b0c3.png