【675】PyTorch —— 图像分类器

PyTorch —— 图像分类器

训练一个图像分类器

我们将按次序的做如下几步：

使用torchvision加载并且归一化CIFAR10的训练和测试数据集
定义一个卷积神经网络
定义一个损失函数
在训练样本数据上训练网络
在测试样本数据上测试网络

1. 数据预处理

　　就是将输入数据处理为可以直接输入模型的类型，参考如下：

import torch
import torchvision
import torchvision.transforms as transforms
 
transform = transforms.Compose(
    [transforms.ToTensor(),
     transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
 
trainset = torchvision.datasets.CIFAR10(root='./data', train=True,
                                        download=True, transform=transform)
trainloader = torch.utils.data.DataLoader(trainset, batch_size=4,
                                          shuffle=True, num_workers=2)
 
testset = torchvision.datasets.CIFAR10(root='./data', train=False,
                                       download=True, transform=transform)
testloader = torch.utils.data.DataLoader(testset, batch_size=4,
                                         shuffle=False, num_workers=2)
 
classes = ('plane', 'car', 'bird', 'cat',
           'deer', 'dog', 'frog', 'horse', 'ship', 'truck')

2. 定义一个卷积神经网络

　　就是上一篇博文的内容

　　定义一个卷积神经网络在这之前先从神经网络章节复制神经网络，并修改它为3通道的图片(在此之前它被定义为1通道)

import torch.nn as nn
import torch.nn.functional as F
 
 
class Net(nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        self.conv1 = nn.Conv2d(3, 6, 5)
        self.pool = nn.MaxPool2d(2, 2)
        self.conv2 = nn.Conv2d(6, 16, 5)
        self.fc1 = nn.Linear(16 * 5 * 5, 120)
        self.fc2 = nn.Linear(120, 84)
        self.fc3 = nn.Linear(84, 10)
 
    def forward(self, x):
        x = self.pool(F.relu(self.conv1(x)))
        x = self.pool(F.relu(self.conv2(x)))
        x = x.view(-1, 16 * 5 * 5)
        x = F.relu(self.fc1(x))
        x = F.relu(self.fc2(x))
        x = self.fc3(x)
        return x
 
 
net = Net()

3. 定义损失函数和优化器

　　定义一个损失函数和优化器让我们使用分类交叉熵Cross-Entropy 作损失函数，动量SGD做优化器。

import torch.optim as optim
 
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)

4. 在训练样本数据上训练网络

　　训练网络这里事情开始变得有趣，我们只需要在数据迭代器上循环传给网络和优化器输入就可以。

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
        inputs, labels = data
 
        # zero the parameter gradients
        optimizer.zero_grad()
 
        # forward + backward + optimize
        outputs = net(inputs)
        loss = criterion(outputs, labels)
        loss.backward()
        optimizer.step()
 
        # print statistics
        running_loss += loss.item()
        if i % 2000 == 1999:    # print every 2000 mini-batches
            print('[%d, %5d] loss: %.3f' %
                  (epoch + 1, i + 1, running_loss / 2000))
            running_loss = 0.0
 
print('Finished Training')

5. 在测试样本数据上测试网络

　　让我们看看网络在整个数据集上的表现。

correct = 0
total = 0
with torch.no_grad():
    for data in testloader:
        images, labels = data
        outputs = net(images)
        _, predicted = torch.max(outputs.data, 1)
        total += labels.size(0)
        correct += (predicted == labels).sum().item()
 
print('Accuracy of the network on the 10000 test images: %d %%' % (
    100 * correct / total))