LeNet网络模型——CIFAR-10数据集进行分类

CIFAR-10数据集由10个类的60000个32*32彩色图像组成，每个类由6000个图像。其中由50000个训练图像和10000个测试图像组成。

数据集分为五个训练批次和一个测试批次，下面采用卷积神经网络对数据集进行分类。

model.py

import torch.nn as nn
import torch.nn.functional as F

"""
pytorch Tensor的通道排序：[batch,channel,height,width]
经过卷积后的尺寸大小计算公式：
N=(W-F+2P)/S + 1
(1)图片大小：w*w;(2)卷积核大小：F*F；（3）步长：s；（4）padding
"""
class LeNet(nn.Module):
    def __init__(self):
        super(LeNet, self).__init__()
        self.conv1 = nn.Conv2d(3,16,5)
        self.pool1 = nn.MaxPool2d(2,2)
        self.conv2 = nn.Conv2d(16,32,5)
        self.pool2 = nn.MaxPool2d(2,2)
        self.fc1 = nn.Linear(32*5*5,120)
        self.fc2 = nn.Linear(120,84)
        self.fc3 = nn.Linear(84,10) # 这次使用的训练集是一个只有十个分类的 分类任务  所以这次就是10

    def forward(self,x):
        x = F.relu(self.conv1(x)) # input(3,32,32) output(16,28,28)
        x = self.pool1(x)         # output(16，14，14)经过最大池化高度和宽度减为原来一半
        x = F.relu(self.conv2(x)) # output(32，10，10)
        x = self.pool2(x)         # output(32，5，5)
        x = x.view(-1,32*5*5)     # 进行展平操作 output(32*5*5)
        x = F.relu(self.fc1(x))   #output(120)
        x = F.relu(self.fc2(x))   #output(84)
        x = self.fc3(x)           #output(10)
        return x

train.py

import torch
import torchvision
import torch.nn as nn
from model import LeNet
import torch.optim as optim
import torchvision.transforms as transforms


def main():
    # transform() 对图像进行预处理的函数
    transform = transforms.Compose(
        [transforms.ToTensor(),
         transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])

    # 50000张训练图片
    # 第一次使用时要将download设置为True才会自动去下载数据集
    train_set = torchvision.datasets.CIFAR10(root='./cifar10-data', train=True,
                                             download=False, transform=transform)
    train_loader = torch.utils.data.DataLoader(train_set, batch_size=36,
                                               shuffle=False, num_workers=0)

    # 10000张验证图片
    # 第一次使用时要将download设置为True才会自动去下载数据集
    val_set = torchvision.datasets.CIFAR10(root='./cifar10-data', train=False,
                                           download=False, transform=transform)
    val_loader = torch.utils.data.DataLoader(val_set, batch_size=5000,
                                             shuffle=False, num_workers=0)

    val_data_iter = iter(val_loader)  # iter 是转化为一个可以迭代的迭代器
    val_image, val_label = val_data_iter.next()

    # classes = ('plane', 'car', 'bird', 'cat',
    #            'deer', 'dog', 'frog', 'horse', 'ship', 'truck')
    device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
    net = LeNet()
    net = net.to(device)
    loss_function = nn.CrossEntropyLoss()
    optimizer = optim.Adam(net.parameters(), lr=0.001)  # 使用Adam优化器

    """
    标准化：output = (input-0.5)/0.5
    反标准化：input = output*0.5+0.5=output/2+0.5
    """

    for epoch in range(5):  # loop over the dataset multiple times；训练5轮

        running_loss = 0.0  # 累加训练过程中的损失
        for step, data in enumerate(train_loader, start=0):
            # 不仅会返回data，还会返回data所对应的步数。
            # get the inputs; data is a list of [inputs, labels]
            inputs, labels = data
            inputs, labels = inputs.to(device), labels.to(device)

            # zero the parameter gradients
            optimizer.zero_grad()
            """
            为什么每次计算一个batch，就需要调用一次optimizer.zero_grad()？
            如果不清除历史梯度，就会对计算的历史梯度进行累加，
            """
            # forward + backward + optimize
            outputs = net(inputs)
            loss = loss_function(outputs, labels)
            loss.backward()
            optimizer.step()

            # print statistics
            running_loss += loss.item()
            if step % 500 == 499:  # print every 500 mini-batches
                # with是一个上下文管理器，with torch.no_grad(): 接下来的计算中不要计算每个节点的误差损失梯度。
                with torch.no_grad():
                    val_image, val_label = val_image.to(device), val_label.to(device)
                    outputs = net(val_image)  # [batch, 10]
                    predict_y = torch.max(outputs, dim=1)[1]
                    # predict_y, val_label).sum() 是一个rensor数据 .item() 获得这个数值
                    accuracy = torch.eq(predict_y, val_label).sum().item() / val_label.size(0)

                    print('[%d, %5d] train_loss: %.3f  test_accuracy: %.3f' %
                          (epoch + 1, step + 1, running_loss / 500, accuracy))
                    running_loss = 0.0

    print('Finished Training')

    save_path = './Lenet.pth'
    torch.save(net.state_dict(), save_path)


if __name__ == '__main__':
    main()

分类训练效果显示：

[1,   500] train_loss: 1.770  test_accuracy: 0.448
[1,  1000] train_loss: 1.452  test_accuracy: 0.515
[2,   500] train_loss: 1.268  test_accuracy: 0.564
[2,  1000] train_loss: 1.172  test_accuracy: 0.597
[3,   500] train_loss: 1.063  test_accuracy: 0.622
[3,  1000] train_loss: 1.008  test_accuracy: 0.635
[4,   500] train_loss: 0.952  test_accuracy: 0.645
[4,  1000] train_loss: 0.910  test_accuracy: 0.649
[5,   500] train_loss: 0.871  test_accuracy: 0.655
[5,  1000] train_loss: 0.839  test_accuracy: 0.668

最好的效果为：66.8%