刘二PyTorch深度学习（十）——CNN进阶

zhaoweiaier

已于 2022-12-20 08:49:20 修改

阅读量577

点赞数 1

文章标签：深度学习 pytorch cnn

于 2022-08-28 22:07:13 首次发布

本文链接：https://blog.csdn.net/zhaoweiaier/article/details/126573102

版权

卷积神经网络（CNN）高级——GoogLeNet

超参数：卷积核的大小就是一个超参数
信息融合：举个例子，就是没门科目的分数*权重1，然后再Σ科目*权重1，即总分，这就是信息融合；说白了就是多个Channel的卷积加起来最后的那个值，就是信息融合

3. 1*1卷积核：最主要的作用是改变通道数，从而减少运算数量（以下图为例：输入的Channel如果是3的话，那么1*1卷积核的Channel也得是3，但是最后输出的Channel却是1，因为要将计算的三个值相加，才能得到右边“x”的值）

计算卷积运算的次数：卷积核宽度的平方*输入图片的宽*输入图片的高*输入通道数*输出通道数
输出图像的通道数=卷积核的个数

Inception块：上图就是一个Inception块，相当于把一个神经网络中所有重复的模块统一封装成一个类，减少代码冗余

执行过程：

第一步：经过卷积层（输入1个通道，输出10个通道，卷积块是5*5），卷积层分三步：卷积-池化-激活（x = F.relu(self.mp(self.conv1(x)))）

第二步：经过Inception块（输入是10个通道，输出是88）

第三步：在经过卷积层（输入88通道，输出20通道，卷积块5*5）

第四步：经过Inception块（输入20，输出88）

第五步：池化（卷积块是2*2）

class Net(torch.nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = torch.nn.Conv2d(1, 10, kernel_size=5)
        self.incep1 = inceptionA(inchannels=10)
        self.conv2 = torch.nn.Conv2d(88, 20, kernel_size=5)
        self.incep2 = inceptionA(inchannels=20)

        self.mp = torch.nn.MaxPool2d(kernel_size=2)

        self.fc=torch.nn.Linear(1408,10)

上面全连接中的1408是通过自动计算得来的，方法如下：

首先，定义forward方法，先不调用全连接层，当调用完x.view方法后，直接调用x.shape就可以自动计算出1408

    def forward(self, x):
        in_size = x.size(0)
        x = F.relu(self.mp(self.conv1(x)))
        x = self.incep1(x)
        x = F.relu(self.mp(self.conv2(x)))
        x = self.incep2(x)
        x = x.view(in_size, -1)
        count = x.shape# 用来计算全连接中的1408
        # x = self.fc(x)

        return x

然后，就是在下面先调用下train方法，就可以自动计算了

代码

import torch
import torch.nn.functional as F
from torchvision import datasets
from torchvision import transforms
from torch.utils.data import DataLoader
import matplotlib.pyplot as plt

batch_size = 64

transform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.1307,), (0.3081,))])

train_datasets = datasets.MNIST(root='./data/mnist/', train=True, download=False, transform=transform)
train_dataloader = DataLoader(dataset=train_datasets, shuffle=True, batch_size=batch_size)

test_datasets = datasets.MNIST(root='./data/mnist/', train=False, download=False, transform=transform)
test_dataloader = DataLoader(dataset=test_datasets, shuffle=False, batch_size=batch_size)


class inceptionA(torch.nn.Module):
    def __init__(self, inchannels):
        super(inceptionA, self).__init__()
        self.branch1x1 = torch.nn.Conv2d(inchannels, 16, kernel_size=1)

        self.branch5x5_1 = torch.nn.Conv2d(inchannels, 16, kernel_size=1)
        self.branch5x5_2 = torch.nn.Conv2d(16, 24, kernel_size=5, padding=2)

        self.branch3x3_1 = torch.nn.Conv2d(inchannels, 16, kernel_size=1)
        self.branch3x3_2 = torch.nn.Conv2d(16, 24, kernel_size=3, padding=1)
        self.branch3x3_3 = torch.nn.Conv2d(24, 24, kernel_size=3, padding=1)

        self.branch_pool = torch.nn.Conv2d(inchannels, 24, kernel_size=1)

    def forward(self, x):
        branch1x1 = self.branch1x1(x)

        branch5x5 = self.branch5x5_1(x)
        branch5x5 = self.branch5x5_2(branch5x5)

        branch3x3 = self.branch3x3_1(x)
        branch3x3 = self.branch3x3_2(branch3x3)
        branch3x3 = self.branch3x3_3(branch3x3)

        branch_pool = F.avg_pool2d(x, kernel_size=3, stride=1, padding=1)
        branch_pool = self.branch_pool(branch_pool)

        out_puts = [branch1x1, branch5x5, branch3x3, branch_pool]

        return torch.cat(out_puts, dim=1)  # Concatenate


class Net(torch.nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = torch.nn.Conv2d(1, 10, kernel_size=5)
        self.incep1 = inceptionA(inchannels=10)
        self.conv2 = torch.nn.Conv2d(88, 20, kernel_size=5)
        self.incep2 = inceptionA(inchannels=20)

        self.mp = torch.nn.MaxPool2d(kernel_size=2)

        self.fc = torch.nn.Linear(1408, 10)

    def forward(self, x):
        in_size = x.size(0)
        x = F.relu(self.mp(self.conv1(x)))
        x = self.incep1(x)
        x = F.relu(self.mp(self.conv2(x)))
        x = self.incep2(x)
        x = x.view(in_size, -1)
        # count = x.shape# 用来计算全连接中的1408
        x = self.fc(x)

        return x


model = Net()

criterion = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(model.parameters(), lr=0.01, momentum=0.05)


def train(epoch):
    running_loss = 0.0
    for batch_index, data in enumerate(train_dataloader, 0):
        inputs, target = data
        outputs = model(inputs)
        loss = criterion(outputs, target)

        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
        running_loss += loss.item()
        if batch_index % 300 == 299:
            print('[%d,%5d] loss: %.3f' % (epoch + 1, batch_index + 1, running_loss / 300))
            running_loss = 0.0


def test():
    correct = 0
    total = 0
    with torch.no_grad():
        for data in test_dataloader:
            images, labels = data
            outputs = model(images)
            # 求出每一行最大值的下标
            _, predicted = torch.max(outputs.data, dim=1)  # dim=0表示行（矩阵从上到下），dim=1表示列（矩阵从左到右）
            total += labels.size(0)  # labels是一个（N,1）的元组
            correct += (predicted == labels).sum().item()

    print('Accuracy on testset %d %%' % (100 * correct / total))
    return correct / total


if __name__ == '__main__':
    loss_list = []
    epoch_list = []
    for epoch in range(10):
        train(epoch)
        loss = test()
        epoch_list.append(epoch)
        loss_list.append(loss)
plt.plot(epoch_list, loss_list)
plt.xlabel('epoch')
plt.ylabel('loss')
plt.show()

Residual Net——ResNet（残差网络）的作用：防止梯度消失（Residual也叫跳连接）

梯度消失的原因：如果将某个大小固定的卷积核进行反复迭代，理想状态下是希望训练状态越来越好，但以CIFAR-10的训练过程为例，在第20次的训练效果达到了最高，第56次的训练结果还不如20次的，所以，并不是迭代次数越多，效果越好，这其中可能的原因就是梯度消失。

下图左边的

Residual Block的实现方式：（首先要保证RB中卷积输出图像的大小不变）

注意：上图中的return那里，是先求和，再激活，这和第一个relu那里是不一样的

import torch
from torchvision import datasets
from torchvision import transforms
from torch.utils.data import DataLoader
import torch.nn.functional as F
import matplotlib.pyplot as plt

batch_size = 64

transform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.1307,), (0.3081,))])

train_datasets = datasets.MNIST(root='./data/mnist/', train=True, download=False, transform=transform)
train_dataloader = DataLoader(dataset=train_datasets, shuffle=True, batch_size=batch_size)

test_datasets = datasets.MNIST(root='./data/mnist/', train=False, download=False, transform=transform)
test_dataloader = DataLoader(dataset=test_datasets, shuffle=False, batch_size=batch_size)


# 定义Residual块
class ResidualBlock(torch.nn.Module):
    def __init__(self, channels):
        super(ResidualBlock, self).__init__()
        self.channels = channels
        # 在定义Res模块时，输入和输出的通道要是一样的，而且既然要保证输入输出通道一致，那么就要加padding了
        self.conv1 = torch.nn.Conv2d(channels, channels, kernel_size=3, padding=1)
        self.conv2 = torch.nn.Conv2d(channels, channels, kernel_size=3, padding=1)

    def forward(self, x):
        y = F.relu(self.conv1(x))
        y = self.conv2(y)
        return F.relu(x + y)


class Net(torch.nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = torch.nn.Conv2d(1, 16, kernel_size=5)
        self.resblock1 = ResidualBlock(channels=16)
        self.conv2 = torch.nn.Conv2d(16, 32, kernel_size=5)
        self.resblock2 = ResidualBlock(channels=32)

        self.mp = torch.nn.MaxPool2d(2)

        self.fc = torch.nn.Linear(512, 10)

    def forward(self, x):
        in_size = x.size(0)
        x = self.mp(F.relu(self.conv1(x)))
        x = self.resblock1(x)

        x = self.mp(F.relu(self.conv2(x)))
        x = self.resblock2(x)

        x = x.view(in_size, -1)
        x = self.fc(x)
        return x


model = Net()

criterion = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(model.parameters(), lr=0.01, momentum=0.05)


def train(epoch):
    running_loss = 0.0
    for batch_index, data in enumerate(train_dataloader, 0):
        inputs, target = data
        outputs = model(inputs)
        loss = criterion(outputs, target)

        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
        running_loss += loss.item()
        if batch_index % 300 == 299:
            print('[%d,%5d] loss: %.3f' % (epoch + 1, batch_index + 1, running_loss / 300))
            running_loss = 0.0


def test():
    correct = 0
    total = 0
    with torch.no_grad():
        for data in test_dataloader:
            images, labels = data
            outputs = model(images)
            # 求出每一行最大值的下标
            _, predicted = torch.max(outputs.data, dim=1)  # dim=0表示行（矩阵从上到下），dim=1表示列（矩阵从左到右）
            total += labels.size(0)  # labels是一个（N,1）的元组
            correct += (predicted == labels).sum().item()

    print('Accuracy on testset %d %%' % (100 * correct / total))
    return correct / total


if __name__ == '__main__':
    loss_list = []
    epoch_list = []
    for epoch in range(10):
        train(epoch)
        loss = test()
        epoch_list.append(epoch)
        loss_list.append(loss)
    plt.plot(epoch_list, loss_list)
    plt.xlabel('epoch')
    plt.ylabel('Accuracy')
    plt.show()