GoogLenet—Inception Module的 PyTorch 实现

以Inception Module实现手写数字识别为例：

import torch
import torch.nn as nn
from torch.utils.data import DataLoader
from torchvision import transforms
from torchvision import datasets  # 放置了许多常用数据集,包括手写数字识别
import torch.nn.functional as F

数据预处理

transform = transforms.Compose([
    transforms.ToTensor(),  # 转张量，将值缩放到[0,1]之间
    transforms.Normalize((0.1307,),(0.3081,))  # 归一化，第一个为均值，第二个为方差
])
# 加载数据
train_dataset = datasets.MNIST(root= "E:/MNIST/mnist",
                              train=True,  # 下载训练集
                              transform=transform,  # 转张量，将值缩放到[0,1]之间.也可以写成transform = transforms.ToTensor()      
                              download=True
                              )

test_dataset = datasets.MNIST(root= "E:/MNIST/mnist",
                              train=False,  # 下载训练集
                              transform=transform,  # 转张量，将值缩放到[0,1]之间
                              download=True
                              )

train_loader = DataLoader(dataset=train_dataset,
                         batch_size=100,
                         shuffle=True)
test_loader = DataLoader(dataset=test_dataset,
                         batch_size=100,
                         shuffle=False)

Inception Block

class InceptionA(nn.Module):
    def __init__(self, in_channels):
        super(InceptionA,self).__init__()
        # 第二个分支
        self.branch1_1 = nn.Conv2d(in_channels, 16, kernel_size=1)
        # 第三个分支
        self.branch5_5_1 = nn.Conv2d(in_channels, 16, kernel_size=1)
        self.branch5_5_2 = nn.Conv2d(16, 24, kernel_size=5, padding=2)
        # 第四个分支
        self.branch3_3_1 = nn.Conv2d(in_channels, 16, kernel_size=1)
        self.branch3_3_2 = nn.Conv2d(16, 24, kernel_size=3, padding=1)
        self.branch3_3_3 = nn.Conv2d(24, 24, kernel_size=3, padding=1)
        # 第一个分支
        self.branch_pool = nn.Conv2d(in_channels, 24, kernel_size=1)
    def forward(self, x):
#         x = self.branch1×1(x)
        
#         x = self.branch5×5_1(x)
#         x = self.branch5×5_2(x)
        # 以上两种写法是错误的，因为等式左边都是 x，而根据图片可知，各分支之间是并联结构。
        branch1_1 = self.branch1_1(x)
        
        branch5_5 = self.branch5_5_1(x)
        branch5_5 = self.branch5_5_2(branch5_5)
        
        branch3_3 = self.branch3_3_1(x)
        branch3_3 = self.branch3_3_2(branch3_3)
        branch3_3 = self.branch3_3_3(branch3_3)
        
        branch_pool = F.avg_pool2d(x, kernel_size=3, stride=1, padding=1)
        branch_pool = self.branch_pool(branch_pool)
        
        outputs = [branch1_1, branch5_5, branch3_3, branch_pool]
        return torch.cat(outputs, dim=1) # (b, c, w, h),则dim=1，即按照通道进行拼接。

搭建模型

# 此模型用到到两个Inception Block
class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(1, 10, kernel_size=5)
        self.conv2 = nn.Conv2d(88, 20, kernel_size=5)
        
        self.incep1 = InceptionA(in_channels=10)
        self.incep2 = InceptionA(in_channels=20)
        
        self.mp = nn.MaxPool2d(2)
        self.fc = nn.Linear(1408, 10)
        
    def forward(self, x):
        in_size = x.size(0)
        x = F.relu(self.mp((self.conv1(x)))) # 卷积、池化、激活，出来之后通道变为了10
        x = self.incep1(x)  # InceptionA输出是通道数并联 = 88
        x = F.relu(self.mp((self.conv2(x))))
        x = self.incep2(x)  # 输入通道=20，输出通道=88
        x = x.view(in_size, -1)
        x = self.fc(x)
        return x
            

model = Net()
loss_fn = torch.nn.CrossEntropyLoss()
optimizer = torch.optim.SGD(model.parameters(), lr= 0.01, momentum= 0.5)

训练函数

def train(epoch):
    runing_loss = 0.0
    for batch_idx, data in enumerate(train_loader, 0):
        inputs, target =data
        optimizer.zero_grad()
        
        outputs = model(inputs)
        loss = loss_fn(outputs, target)
        loss.backward()
        optimizer.step()
        
        runing_loss += loss.item()
        if batch_idx % 300  == 299:
            print("[%d, %5d] loss: %.3f" % (epoch +1, batch_idx+1, runing_loss/300))
            runing_loss = 0.0

测试函数

def test():
    correct = 0
    total =0
    with torch.no_grad():
        for data in test_loader:
            images, labels =data
            outputs = model(images)  
            _, predicted = torch.max(outputs.data, dim =1 )  # 返回两个值，第一个是最大值，第二个是最大值的索引。dim=1表示在列维度求以上结果，dim = 0表示在行维度求以上结果。         
            total += labels.size(0)  # 每一个batch_size 中labels是一个（N，1）的元组，size(0)=N
            correct +=(predicted == labels).sum().item()  # 对的总个数
    print("Accuracy on the test set %s %%" % (100*correct/total))

网络启动

if __name__=="__main__":
    for epoch in range(10):
        train(epoch)
        if epoch % 2 ==0:
            test()

[1,   300] loss: 0.806
[1,   600] loss: 0.189
Accuracy on the test set 96.0 %
[2,   300] loss: 0.125