【戴口罩高精度识别项目】神经网络搭建

weixin_45056254

已于 2023-02-27 14:23:29 修改

阅读量215

点赞数

文章标签：人工智能

于 2023-02-26 17:29:18 首次发布

本文链接：https://blog.youkuaiyun.com/weixin_45056254/article/details/129226243

版权

该文展示了如何使用Pytorch搭建一个4层的Resnet神经网络，输入为224x224x3的RGB图像，输出为1000类。网络包括layer0至layer4，每个layer由ResBlock组成，涉及卷积、批量归一化和激活函数。最后，网络在CIFAR10数据集上进行训练。

摘要生成于 C知道，由 DeepSeek-R1 满血版支持，前往体验 >

1.Pytorch搭建4层Resnet神经网络，基本参数如下：

1）一开始的输入大小设置为:(224,224,3)

2）layer0层有1层卷积、BN、激活、池化两个部分，卷积核大小为7X7，通道为64，步长为2，池化层大小为 3x3，步长为2；

3）layer1-layer4每层都有ResBlock1、ResBlock2，每个模块都经过两次2维卷积，每个模块可能会经历下采样与尺寸通道的变化，具体见layer1-layer4架构细节图ResBlock1、ResBlock2每个模块都会经历卷积、BN算法(解决层数多了后收敛慢的问题)、激活的过程，具体见各层关系对应图

各层对应关系：

import torch
import torch.nn as nn
from torchsummary import summary
import torchvision
import torchvision.transforms as transforms

"""目的：搭建一个4层残差网络：输入：224x224x3 RGB彩图,输出：1000类"""
class Resnet18(nn.Module):
    #每个神经网络的构成
    def __init__(self,num_classes):
        super().__init__()#继承父类的方法，即使定义了def __init__(self,num_classes):也不会覆盖父类的def __init__(self,num_classes)函数
        channels_list = [64,128,256,512]#为方便预先定义每层通道数量
        #layer 0
        self.layer_0 = nn.Sequential(       
            nn.Conv2d(in_channels=3,out_channels=64,kernel_size=7,stride=2,padding=3),
            nn.BatchNorm2d(64),
            nn.ReLU(),
            nn.MaxPool2d(kernel_size=3,stride=2,padding=1),
        ) 
        # layer 1
        self.layer_1 = nn.Sequential(
            ResBlock(False,channels_list[0],channels_list[0]),#不做下采样         
            ResBlock(False,channels_list[0],channels_list[0]),#不做下采样             
        )
        # layer 2        
        self.layer_2 = nn.Sequential(
            ResBlock(True,channels_list[0],channels_list[1]),#做下采样         
            ResBlock(False,channels_list[1],channels_list[1]),#不做下采样             
        )        
        # layer 3
        self.layer_3 = nn.Sequential(
            ResBlock(True,channels_list[1],channels_list[2]),#做下采样          
            ResBlock(False,channels_list[2],channels_list[2]), #不做下采样           
        )          
        # layer 4
        self.layer_4 = nn.Sequential(
            ResBlock(True,channels_list[2],channels_list[3]),#做下采样            
            ResBlock(False,channels_list[3],channels_list[3]),#不做下采样           
        )  
        #layer 5
        self.aap = nn.AdaptiveAvgPool2d((1, 1))# 平均池化
        self.flatten = nn.Flatten(start_dim=1)# 拉直
        self.fc = nn.Linear(channels_list[3],num_classes)# 全连接层
        
    #每层神经网络之间的链接
    def forward(self,x):
        x = self.layer_0(x)
        x = self.layer_1(x)
        x = self.layer_2(x)
        x = self.layer_3(x)
        x = self.layer_4(x)  
        x = self.aap(x)
        x = self.flatten(x)
        x = self.fc(x)
        return x
class ResBlock(nn.Module):
    def __init__(self,down_sample,in_channels,out_channels):
        super().__init__()
        if down_sample:
            self.conv1 = nn.Conv2d(in_channels,out_channels,3,2,1)
            self.bn1 = nn.BatchNorm2d(out_channels)
            self.relu1 = nn.ReLU()
            
            self.conv2 = nn.Conv2d(out_channels,out_channels,3,1,1)
            self.bn2 = nn.BatchNorm2d(out_channels)
            self.relu2 = nn.ReLU()
            self.shortcut = nn.Sequential(
                nn.Conv2d(in_channels,out_channels,1,2,0),
                nn.BatchNorm2d(out_channels))#Sequential不是一个空壳，加了nn.Conv2d、nn.BatchNorm2d规则，所以shortcut接收任何数都是先卷积再BN算法
            self.relu3 = nn.ReLU()#一个模块的最终结果也是需要被激活一下的
        else:
            self.conv1 = nn.Conv2d(in_channels,out_channels,3,1,1)
            self.bn1 = nn.BatchNorm2d(out_channels)
            self.relu1 = nn.ReLU()
            
            self.conv2 = nn.Conv2d(out_channels,out_channels,3,1,1)
            self.bn2 = nn.BatchNorm2d(out_channels)
            self.relu2 = nn.ReLU()
            self.shortcut = nn.Sequential()#Sequential是一个空壳，没有规则，所以shortcut接收任何数都是没有规则的出去
            self.relu3 = nn.ReLU()#一个模块的最终结果也是需要被激活一下的
    def forward(self,x):
        shortcut = self.shortcut(x)#注意这里做下采样和不做下采样的self.shortcut对x的处理结果是不一样的
        #ResBlock的两次卷积
        x= self.conv1(x)
        x = self.bn1(x)
        x = self.relu1(x)
            
        x= self.conv2(x)
        x = self.bn2(x)
        x = self.relu2(x) 
        #残差连接
        x = x + shortcut
        x = self.relu3(x)   
        return x

resnet18 =Resnet18(10)
#resnet18 =Resnet18(10).to('cuda:0')

"""加载模型"""
transform = transforms.Compose(
    [transforms.ToTensor(),
     transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])

batch_size = 4

trainset = torchvision.datasets.CIFAR10(root='./1', train=True,
                                        download=True, transform=transform)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=batch_size,
                                          shuffle=True, num_workers=2)

testset = torchvision.datasets.CIFAR10(root='./1', train=False,
                                       download=True, transform=transform)
testloader = torch.utils.data.DataLoader(testset, batch_size=batch_size,
                                         shuffle=False, num_workers=2)

optimizer = torch.optim.SGD(resnet18.parameters(),lr=0.01)# 加载res18模型及res18模型的优化器
criterion = nn.CrossEntropyLoss()#加载res18模型LOSS函数

for epoch in range(2):  # loop over the dataset multiple times

    running_loss = 0.0
    for i, data in enumerate(trainloader, 0):
        # get the inputs; data is a list of [inputs, labels]
        inputs, labels = data
        #inputs.to('cuda:0')
        # zero the parameter gradients
        optimizer.zero_grad()

        # forward + backward + optimize
        outputs = resnet18(inputs)
        loss = criterion(outputs, labels)
        loss.backward()
        optimizer.step()

        # print statistics
        running_loss += loss.item()
        print(loss.item())
        if i % 2000 == 1999:    # print every 2000 mini-batches
            print(f'[{epoch + 1}, {i + 1:5d}] loss: {running_loss / 2000:.3f}')
            running_loss = 0.0

print('Finished Training')

Gitee链接：✉