经典卷积神经网络 - VGG

使用块的网络 - VGG。

使用多个$3\times 3$的要比使用少个$5\times 5$的效果要好。


VGG全称是Visual Geometry Group，因为是由Oxford的Visual Geometry Group提出的。AlexNet问世之后，很多学者通过改进AlexNet的网络结构来提高自己的准确率，主要有两个方向：小卷积核和多尺度。而VGG的作者们则选择了另外一个方向，即加深网络深度。

网络架构

卷积网络的输入是224 * 224的RGB图像，整个网络的组成是非常格式化的，基本上都用的是3 * 3的卷积核以及 2 * 2的max pooling，少部分网络加入了1 * 1的卷积核。因为想要体现出“上下左右中”的概念，3*3的卷积核已经是最小的尺寸了。

VGG16相比之前网络的改进是3个33卷积核来代替7x7卷积核，2个33卷积核来代替5*5卷积核，这样做的主要目的是在保证具有相同感知野的条件下，减少参数，提升了网络的深度。

多个VGG块后接全连接层。

不同次数的重复块得到不同的架构，如VGG-16，VGG-19等。

VGG：更大更深的AlexNet。

总结：

• VGG使用可重复使用的卷积块来构建深度卷积神经网络
• 不同的卷积块个数和超参数可以得到不同复杂度的变种

代码实现

使用数据集CIFAR

model.py

"""
@Author  :shw
@Date    :2023/10/23 16:35
"""
from torch import nn

# 输入图像大小 224x224x3
class VGG(nn.Module):
    def __init__(self, *args, **kwargs) -> None:
        super().__init__(*args, **kwargs)
        self.features = nn.Sequential(
            nn.Conv2d(3,64,kernel_size=3,padding=1),
            nn.BatchNorm2d(64),  # 出现了梯度消失问题，所以使用了BN层
            nn.ReLU(inplace=True),
            nn.Conv2d(64,64,kernel_size=3,padding=1),
            nn.BatchNorm2d(64),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=2,stride=2),

            nn.Conv2d(64,128,kernel_size=3,padding=1),
            nn.BatchNorm2d(128),
            nn.ReLU(inplace=True),
            nn.Conv2d(128,128,kernel_size=3,padding=1),
            nn.BatchNorm2d(128),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=2,stride=2),

            nn.Conv2d(128,256,kernel_size=3,padding=1),
            nn.BatchNorm2d(256),
            nn.ReLU(inplace=True),
            nn.Conv2d(256,256,kernel_size=3,padding=1),
            nn.BatchNorm2d(256),
            nn.ReLU(inplace=True),
            nn.Conv2d(256,256,kernel_size=3,padding=1),
            nn.BatchNorm2d(256),
            nn.ReLU(inplace=True),
            nn.Conv2d(256, 256, kernel_size=3, padding=1),
            nn.BatchNorm2d(256),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=2,stride=2),

            nn.Conv2d(256,512,kernel_size=3,padding=1),
            nn.BatchNorm2d(512),
            nn.ReLU(inplace=True),
            nn.Conv2d(512,512,kernel_size=3,padding=1),
            nn.BatchNorm2d(512),
            nn.ReLU(inplace=True),
            nn.Conv2d(512, 512, kernel_size=3, padding=1),
            nn.BatchNorm2d(512),
            nn.ReLU(inplace=True),
            nn.Conv2d(512,512,kernel_size=3,padding=1),
            nn.BatchNorm2d(512),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=2,stride=2),

            nn.Conv2d(512, 512,kernel_size=3,padding=1),
            nn.BatchNorm2d(512),
            nn.ReLU(inplace=True),
            nn.Conv2d(512, 512,kernel_size=3,padding=1),
            nn.BatchNorm2d(512),
            nn.ReLU(inplace=True),
            nn.Conv2d(512, 512,kernel_size=3,padding=1),
            nn.BatchNorm2d(512),
            nn.ReLU(inplace=True),
            nn.Conv2d(512, 512, kernel_size=3,padding=1),
            nn.BatchNorm2d(512),
            nn.ReLU(inplace=True),
            nn.MaxPool2d(kernel_size=2,stride=2)
        )
        self.avgpool = nn.AdaptiveAvgPool2d((7, 7))
        self.classifier = nn.Sequential(
            nn.Flatten(),
            nn.Linear(7*7*512,4096),
            nn.ReLU(inplace=True),
            nn.Dropout(),

            nn.Linear(4096, 4096),
            nn.ReLU(inplace=True),
            nn.Dropout(),
            nn.Linear(4096, 10),
            # 使用了交叉熵损失函数后，最后一层不再需要使用softmax
            # nn.Softmax()
        )

    def forward(self,x):
        x = self.features(x)
        x = self.avgpool(x)
        x = self.classifier(x)
        return x

train.py

from torchvision import transforms, datasets
from torch.utils.data import DataLoader
from torch.utils.tensorboard import SummaryWriter
from torch import nn
import torch
from model import VGG


class Train:
    def __init__(self):
        self.epochs = 10
        self.lr = 1e-4  # 1e-3 -> 1e-4
        self.train_step = 0
        self.test_step = 0
        self.batch_size = 64  # 32->64
        self.train_size = None
        self.test_size = None
        self.train_loader = None
        self.test_loader = None

    def get_data(self):
        transformer = transforms.Compose([
            transforms.Resize((224, 224)),
            transforms.ToTensor()
        ])
        train_data = datasets.CIFAR10(root="./dataset", train=True, transform=transformer, download=True)
        test_data = datasets.CIFAR10(root="./dataset", train=False, transform=transformer, download=True)
        self.train_size = len(train_data)
        self.test_size = len(test_data)
        self.train_loader = DataLoader(train_data, batch_size=self.batch_size, shuffle=True, num_workers=0)
        self.test_loader = DataLoader(test_data, batch_size=self.batch_size, shuffle=True, num_workers=0)
        print("训练集大小：{}".format(self.train_size))
        print("测试集大小：{}".format(self.test_size))

    def get_accuracy(self, accuracy_num, size):
        return accuracy_num / size

    def train(self):

        self.get_data()

        device = torch.device("cpu")
        if torch.backends.mps.is_available():
            device = torch.device("mps")
        elif torch.cuda.is_available():
            device = torch.device("cuda")

        vgg = VGG()
        vgg = vgg.to(device)

        # 定义损失函数
        loss_func = nn.CrossEntropyLoss()
        loss_func = loss_func.to(device)
        # 定义优化器
        # optimizer = torch.optim.SGD(vgg.parameters(), lr=self.lr, momentum=0.9)
        optimizer = torch.optim.Adam(vgg.parameters(),lr=self.lr,betas=(0.9,0.999),eps=1e-08)

        writer = {
            'train': SummaryWriter(log_dir="./logs/t2/train"),
            'test': SummaryWriter(log_dir="./logs/t2/test")
        }

        for epoch in range(self.epochs):
            print("-------------------第 {} 个Epoch训练开始-------------------".format(epoch + 1))
            total_train_accuracy_num = 0
            total_test_accuracy_num = 0
            total_test_loss = 0
            total_train_loss = 0

            vgg.train()
            for data in self.train_loader:
                self.train_step += 1
                images, targets = data
                images = images.to(device)
                targets = targets.to(device)

                outputs = vgg(images)
                loss_out = loss_func(outputs, targets)
                vgg.zero_grad()
                loss_out.backward()
                optimizer.step()

                if self.train_step % 100 == 0:
                    print("训练次数：{}，Loss：{}".format(self.train_step, loss_out))
                    writer['train'].add_scalar('Train Loss', loss_out, self.train_step)
                total_train_accuracy_num += (torch.argmax(outputs,dim=1) == targets).sum()
                total_train_loss += loss_out
            train_accuracy =self.get_accuracy(total_train_accuracy_num , self.train_size)
            writer['train'].add_scalar("Accuracy", train_accuracy, epoch + 1)
            # print("Epoch：{}，Train Accuracy：{}".format(epoch+1,train_accuracy))
            # print("Epoch：{}，Total Train Loss：{}".format(epoch+1,total_train_loss))

            vgg.eval()
            with torch.no_grad():
                for data in self.test_loader:
                    self.test_step += 1
                    images, targets = data
                    images = images.to(device)
                    targets = targets.to(device)
                    outputs = vgg(images)
                    loss_out = loss_func(outputs, targets)
                    total_test_loss += loss_out
                    total_test_accuracy_num += (torch.argmax(outputs,dim=1) == targets).sum()
            test_accuracy = self.get_accuracy(total_test_accuracy_num, self.test_size)
            # print("Epoch：{}，Test Accuracy：{}".format(epoch + 1, test_accuracy))
            # print("Epoch：{}，Total Test Loss：{}".format(epoch + 1, total_test_loss))
            print("Epoch：{}，Train Loss：{}，Test Loss：{}，Train Accuracy：{}，Test Accuracy：{}".format(epoch + 1,
                                                                                                  total_train_loss,
                                                                                                  total_test_loss,
                                                                                                  train_accuracy,
                                                                                                  test_accuracy))
            writer['test'].add_scalar("Test Total Loss", total_test_loss, epoch + 1)
            writer['test'].add_scalar("Accuracy", test_accuracy, epoch + 1)

            torch.save(vgg,"./model/vgg_{}.pth".format(epoch+1))
        writer['test'].close()
        writer['train'].close()

main.py

from train import Train

if __name__ == '__main__':
    train = Train()
    train.train()