PyTorch深度学习实践——卷积神经网络（GoogLeNet部分实现、ResNet )

参考资料

参考资料1：https://blog.csdn.net/bit452/article/details/109693790
参考资料2：http://biranda.top/Pytorch%E5%AD%A6%E4%B9%A0%E7%AC%94%E8%AE%B0012%E2%80%94%E2%80%94Advancedd_CNN/

torch.nn.Conv2d 为什么只定义卷积核的大小，而不定义卷积核的具体数值

卷积核都是随机的
https://segmentfault.com/q/1010000022234007

GoogLeNet部分实现

1×1Conv下面的括号是输出有几个通道。


每次维度为[64,1,28,28]的图像，经过两次Inception Module 后，再通过一个全连接 生成10个概率。

import torch
import torch.nn as nn
from torchvision import transforms
from torchvision import datasets
from torch.utils.data import DataLoader
import torch.nn.functional as F
import torch.optim as optim

# prepare dataset
batch_size = 64
transform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.1307,), (0.3081,))])  # 归一化,均值和方差

train_dataset = datasets.MNIST(root='./dataset/mnist/', train=True, download=True, transform=transform)
train_loader = DataLoader(train_dataset, shuffle=True, batch_size=batch_size)
test_dataset = datasets.MNIST(root='./dataset/mnist/', train=False, download=True, transform=transform)
test_loader = DataLoader(test_dataset, shuffle=False, batch_size=batch_size)

# design model using class
class InceptionA(nn.Module):
    # 仅是一个模块，其中的输入通道数并不能够指明
    def __init__(self, in_channels):
        super(InceptionA, self).__init__()
        # 1
        # 定义一个输出通道为16的单一的1×1的卷积
        self.branch1x1 = nn.Conv2d(in_channels, 16, kernel_size=1)
        # 2
        self.branch5x5_1 = nn.Conv2d(in_channels, 16, kernel_size=1)
        self.branch5x5_2 = nn.Conv2d(16, 24, kernel_size=5, padding=2)
        # 3
        self.branch3x3_1 = nn.Conv2d(in_channels, 16, kernel_size=1)
        self.branch3x3_2 = nn.Conv2d(16, 24, kernel_size=3, padding=1)
        self.branch3x3_3 = nn.Conv2d(24, 24, kernel_size=3, padding=1)
        # 4
        # init内定义1x1卷积（输入通道 输出通道 卷积核大小）
        self.branch_pool = nn.Conv2d(in_channels, 24, kernel_size=1)

    def forward(self, x):
        # 1
        branch1x1 = self.branch1x1(x)
        # 2
        branch5x5 = self.branch5x5_1(x)
        branch5x5 = self.branch5x5_2(branch5x5)
        # 3
        branch3x3 = self.branch3x3_1(x)
        branch3x3 = self.branch3x3_2(branch3x3)
        branch3x3 = self.branch3x3_3(branch3x3)
        # 4
        # avg_pool2d->均值池化函数 stride以及padding需要手动设置以保持图像的宽度和高度不变，
        #这里设置kernel_size=3, stride=1, padding=1是为了 图像的长和宽不变。
        branch_pool = F.avg_pool2d(x, kernel_size=3, stride=1, padding=1)
        # 括号内branch_pool的是池化后的结果，括号外的branch_pool是定义的1x1卷积，赋值给对象branch_pool
        branch_pool = self.branch_pool(branch_pool)
        # 利用Concatenate按通道维度方向进行拼接可得到输出图像。dim=1 意味着按下标为1的维度方向拼接，在图像有四个维度（B,C,W,H），dim=1的是通道C。
        # cat拼接
        outputs = [branch1x1, branch5x5, branch3x3, branch_pool]
        return torch.cat(outputs, dim=1)  # b,c,w,h  c对应的是dim=1

class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(1, 10, kernel_size=5)
        # 在Inception的定义中，拼接后的输出通道数为24+16+24+24=88个
        self.conv2 = nn.Conv2d(88, 20, kernel_size=5)  # 88 = 24x3 + 16

        self.incep1 = InceptionA(in_channels=10)  # 与conv1 中的10对应
        self.incep2 = InceptionA(in_channels=20)  # 与conv2 中的20对应

        self.mp = nn.MaxPool2d(2)
        # 关于1408：
        # 每次卷积核是5x5，则卷积后原28x28的图像变为24x24的
        # 再经过最大池化，变为12x12的
        # 以此类推最终得到4x4的图像，又inception输出通道88，则转为一维后为88x4x4=1408个
        self.fc = nn.Linear(1408, 10)

    def forward(self, x):
        in_size = x.size(0) # in_size:64  , size([64,1,28,28])
        # 1、将一个1通道的转为10通道的，
        x = F.relu(self.mp(self.conv1(x)))
        # 2、将10通道的转为24+16+24+24=88 通道的
        x = self.incep1(x)
        # 3、将88通道的转为20通道的，
        x = F.relu(self.mp(self.conv2(x)))
        # 4、将20通道的转为88通道的，
        x = self.incep2(x)
        # 上面有解释为何1408
        x = x.view(in_size, -1)
        x = self.fc(x)

        return x

model = Net()

#GPU
#cuda 0是选择第一块显卡，cuda 1是选择第二块显卡。
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
#将来model迁移到device
model.to(device)

# construct loss and optimizer
#交叉熵损失函数CrossEntropyLoss()
criterion = torch.nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.5)

# training cycle forward, backward, update
def train(epoch):
    running_loss = 0.0
    # batch_size = 64 ，最前面已经声明了
    for batch_idx, data in enumerate(train_loader, 0):
        inputs, target = data
        #GPU
        inputs, target = inputs.to(device), target.to(device)
        optimizer.zero_grad()# 优化器梯度清零

        outputs = model(inputs)
        loss = criterion(outputs, target)
        loss.backward()
        optimizer.step()

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


def test():
    correct = 0
    total = 0
    with torch.no_grad():# 测试集不需要计算梯度
        for data in test_loader:
            images, labels = data
            #GPU
            # send the images and labels at every step to the GPU。
            images, labels = images.to(device), labels.to(device)

            outputs = model(images)
            _, predicted = torch.max(outputs.data, dim=1)
            total += labels.size(0)
            correct += (predicted == labels).sum().item()
    print('accuracy on test set: %d %% ' % (100 * correct / total))


if __name__ == '__main__':
    for epoch in range(10):
        train(epoch)
        test()

ResNet残差网络

plain net 纯网络
residual net 残差网络

import torch
import torch.nn as nn
from torchvision import transforms
from torchvision import datasets
from torch.utils.data import DataLoader
import torch.nn.functional as F
import torch.optim as optim

# prepare dataset
batch_size = 64
transform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.1307,), (0.3081,))])  # 归一化,均值和方差
#训练集
train_dataset = datasets.MNIST(root='./dataset/mnist/', train=True, download=True, transform=transform)
train_loader = DataLoader(train_dataset, shuffle=True, batch_size=batch_size)
#测试集
test_dataset = datasets.MNIST(root='./dataset/mnist/', train=False, download=True, transform=transform)
test_loader = DataLoader(test_dataset, shuffle=False, batch_size=batch_size)

# design model using class
class ResidualBlock(nn.Module):
    def __init__(self, channels):
        super(ResidualBlock, self).__init__()
        self.channels = channels
        self.conv1 = nn.Conv2d(channels, channels, kernel_size=3, padding=1)
        self.conv2 = 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(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(1, 16, kernel_size=5)
        self.conv2 = nn.Conv2d(16, 32, kernel_size=5)  # 88 = 24x3 + 16

        self.rblock1 = ResidualBlock(16)
        self.rblock2 = ResidualBlock(32)

        self.mp = nn.MaxPool2d(2)
        self.fc = nn.Linear(512, 10)

    def forward(self, x):
        in_size = x.size(0)

        x = self.mp(F.relu(self.conv1(x)))
        x = self.rblock1(x)
        x = self.mp(F.relu(self.conv2(x)))
        x = self.rblock2(x)

        x = x.view(in_size, -1)#(64,32×4×4=512)
        x = self.fc(x)
        return x

model = Net()

#GPU
#cuda 0是选择第一块显卡，cuda 1是选择第二块显卡。
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
#将来model迁移到device
model.to(device)

# construct loss and optimizer
criterion = torch.nn.CrossEntropyLoss()
optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.5)

# training cycle forward, backward, update
def train(epoch):
    running_loss = 0.0
    for batch_idx, data in enumerate(train_loader, 0):
        inputs, target = data
        #GPU
        inputs, target = inputs.to(device), target.to(device)
        optimizer.zero_grad()

        outputs = model(inputs)
        loss = criterion(outputs, target)
        loss.backward()
        optimizer.step()

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

def test():
    correct = 0
    total = 0
    with torch.no_grad():
        for data in test_loader:
            images, labels = data
            # GPU
            # send the images and labels at every step to the GPU。
            images, labels = images.to(device), labels.to(device)

            outputs = model(images)
            _, predicted = torch.max(outputs.data, dim=1)
            total += labels.size(0)
            correct += (predicted == labels).sum().item()
    print('accuracy on test set: %d %% ' % (100 * correct / total))

if __name__ == '__main__':
    for epoch in range(10):
        train(epoch)
        test()