基于torchvision数据集CIFAR10的图像分类器实现

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

device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")


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


# transform.Compose 将图片的多个变换（transforms）组合在一起形成一个transform,参数是一个列表，列表中的元素都是transform
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,
                                        transform=transform)  # 没有下载就将download=True,下载过了就省略，下载CIFAR10训练集到本地
trainloader = torch.utils.data.DataLoader(trainset, batch_size=4, shuffle=True, num_workers=0)  # 将数据集自动打乱分成批量
testset = torchvision.datasets.CIFAR10(root='./data', train=False, transform=transform)
testloader = torch.utils.data.DataLoader(testset, batch_size=4, shuffle=False,
                                         num_workers=0)  # 注意：num_workers=0采用主线程不容易报错，如果多个线程，还要设置线程工作
classes = ('plane', 'car', 'bird', 'cat', 'deer', 'dog', 'frog', 'horse', 'ship', 'truck')
'''
#图片操作
def imshow(img): 
    img = img / 2 + 0.5 # unnormalize
    npimg = img.numpy()
    plt.imshow(np.transpose(npimg, (1, 2, 0)))
    plt.show()
# get some random training images
dataiter = iter(trainloader)   #iter(trainloader) trainloader必须是可迭代对象。返回一个迭代器
images, labels = dataiter.next() # 输出trainloader中第一个batch，也就是4个图像-标签
# show images
imshow(torchvision.utils.make_grid(images)) #torchvision.utils.make_grid()将images中的四幅图片拼接起来
# print labels
print(' '.join('%5s' % classes[labels[j]] for j in range(4)))
'''

net = Net()
net.to(device)
criterion = nn.CrossEntropyLoss()
optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)
# 训练训练集两次
for epoch in range(10):
    running_loss = 0.0
    for i, data in enumerate(trainloader,
                             0):  # enumerate()将trainloader（可遍历对象）在每个元素左边加上一个索引变为一个元组（index,element of trainloader）
        inputs, labels = data  # trainloader有两列，第一列是输入，第二列是label
        inputs, labels = inputs.to(device), labels.to(device)
        optimizer.zero_grad()  # 每次loss计算前要把梯度清零
        outputs = net(inputs)  # inputs输入到网络中
        loss = criterion(outputs, labels)  # outputs 必须在labels左边
        loss.backward()
        optimizer.step()
        running_loss += loss.item()  # loss.item()指的是每一次经过前馈路径得到结果和label的差,相加然后求平均。每两千组求一次loss的平均值。
        if i % 2000 == 1999:
            print('[%d, %5d] loss: %.3f' % (epoch + 1, i + 1, running_loss / 2000))
            running_loss = 0.0

# 在测试集上的准确度
correct = 0
total = 0
with torch.no_grad():  # 在测试集数据上没必要追踪梯度，所以停止梯度追踪
    for inputs, labels in testloader:  # 每批量
        inputs, labels = inputs.to(device), labels.to(device)
        outputs = net(inputs)
        loss = criterion(outputs, labels)  # outputs 必须在labels左边
        _, predicted = torch.max(outputs.data,
                                 1)  # outputs.data是一个张量,（tensor,dim）dim=1是每行的最大值，dim=0是每列的最大值,返回值是两个tensor，第二个tensor（1xbatch_size）是每行最大值对应的索引
        total += labels.size(
            0)  # labels.size(axis) 有axis维的tensor axis=0返回第一维的bound，axis=1返回第二维的bound。一维tensor axis=0返回列的个数
        correct += (
                predicted == labels).sum().item()  # （==）返回一个张量，此张量的size和labels一样，tensor里的值为Ture/False。 tensor.sum()将tensor中所有相加，正确为1，错误为0并且返回一个只有一个值的tensor，只有一个值的tensor.item()返回tensor中值
print('Accuracy of the network on 10000 test images %d %%' % (100 * correct / total))