AlexNet网络的搭建

AlexNet网络解析

亮点

1. 首次引入GPU进行网络加速训练,其中，下图是使用了两块GPU进行训练

   

2. 使用了ReLU激活函数

3. 在全连接层的前两层使用了Dropout随机失活神经元的操作，减少过拟合

网络结构

经卷积后矩阵尺寸大小计算公式为：

N = (W-F+P1+P2)/S+1

W-输入图片的大小W*W;F-卷积核大小FxF;

P1:图片上方padding的个数;P2:图片下方padding的个数;S:步长

Conv1

输入:(3,224,224)

W=224,F=11,P=[1,2],S=4,N=(224-11+1+2)/4+1=55

输出:(96,55,55)

Maxpool1

输入:(96,55,55)

W=55,F=3,P=[0,0],S=2,N=(55-3)/2+1=27

输出:(96,27,27)

Conv2

输入:(96,27,27)

W=27,F=5,P=[2,2],S=1,N=(27-5+4)/1+1=27

输出:(256,27,27)

Maxpool2

输入:(256,27,27)

W=27,F=3,P=[0,0],S=2,N=(27-3)/2+1=13

输出:(256,13,13)

Conv3

输入:(256,13,13)

W=13,F=3,P=[1,1],S=1,N=(13-3+1+1)/1+1=13

输出:(384,13,13)

Conv4

输入:(384,13,13)

W=13,F=3,P=[1,1],S=1,N=(13-3+1+1)/1+1=13

输出:(384,13,13)

Conv5

输入:(384,13,13)

W=13,F=3,P=[1,1],S=1,N=(13-3+1+1)/1+1=13

输出:(256,13,13)

代码复现

class AlexNet(nn.Module):
    def __init__(self,num_classes=1000,init_weights=False):
        super(AlexNet, self).__init__()
        # nn.Sequential         简化步骤
        self.features = nn.Sequential(
            # 特征提取层
            # Input[3,224,224]    output[48,55,55]
            nn.Conv2d(3,48,kernel_size=11,stride=4,padding=2),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=3,stride=2),
            nn.Conv2d(48,128,kernel_size=5,padding=2),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=3,stride=2),
            nn.Conv2d(128,192,kernel_size=3,padding=1),
            nn.ReLU(inplace=True),
            nn.Conv2d(192, 192, kernel_size=3, padding=1),
            nn.ReLU(inplace=True),
            nn.Conv2d(192, 128, kernel_size=3, padding=1),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=3, stride=2),
        )

        self.classifier = nn.Sequential(
            # 全连接层
            nn.Dropout(p=0.5),  # p:神经元随机失活的比例
            nn.Linear(128*6*6,2048),
            nn.ReLU(inplace=True),
            nn.Dropout(p=0.5),
            nn.Linear(2048,2048),
            nn.ReLU(inplace=True),
            nn.Linear(2048,num_classes)  # num_classes：分类的类别数
        )

        if init_weights:
            self._initialize_weights()

    def forward(self,x):
        x = self.features(x)
        x = torch.flatten(x,start_dim=1)
        x = self.classifier(x)
        return x

    # 初始化权重
    def _initialize_weights(self):
        # 遍历整个modules
        for m in self.modules():
            # 发现有nn.Conv2d这个结构
            if isinstance(m,nn.Conv2d):
                # 凯明初始化权重
                nn.init.kaiming_normal_(m.weight,mode='fan_out', nonlinearity='relu')
                if m.bias is not None:
                    nn.init.constant_(m.bias,0)
            # 发现有nn.Linear，则使用正态分布初始化函数
            elif isinstance(m,nn.Linear):
                nn.init.normal_(m.weight,0,0.01)
                nn.init.constant_(m.bias,0)