LeNet神经网络-pytorch实现

LeNet

1.网络架构

抛开SVM支持向量机和MLP多层感知机，接触的第一个CNN网络架构

如图所示可见其结构为：

输入的二维图像处理后，先经过两次卷积层到池化层，再经过全连接层，最后使用softmax分类作为输出层。

2.pytorch网络设计

网络构造代码部分：

class LeNet5(nn.Module):
    def __init__(self):
        super().__init__()
        self.conv1 = nn.Conv2d(1, 6, 5, padding=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 = F.max_pool2d(F.relu(self.conv1(x)), (2, 2))
        x = F.max_pool2d(F.relu(self.conv2(x)), (2, 2))
        x = x.view(x.shape[0], -1)
        x = F.relu(self.fc1(x))
        x = F.relu(self.fc2(x))
        x = self.fc3(x)
        return x

3.网络测试

网络测试部分我使用的是minist数据集，为了贴近真实（主要是方便我自己懂），在下载了数据集之后将其转为了图片数据集，更为直观。数据集分为train 和test两部分，在测试中需要做如下配置：

1.依赖资源引入

draw_tool是一个自己编写的绘制loss，acc的画图库，device使用了我电脑的1050ti显卡。

import torch
from matplotlib import pyplot as plt
import torch.optim as optim
from torch.autograd import Variable
import torch.nn.functional as F
from torch import nn
from torchsummary import summary
from torchvision import transforms
from torch.utils.data import Dataset, DataLoader
from PIL import Image
import draw_tool

root = "F:/pycharm/dataset/mnist/MNIST/"
device = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
draw = draw_tool.draw_tool()

2.数据集的读取和分类

#加载图片
def default_loader(path):
    return Image.open(path).convert('RGB')

#构造标注和图片相关
class MyDataset(Dataset):
    def __init__(self, txt, transform=None, target_transform=None, loader=default_loader):
        fh = open(txt, 'r')
        imgs = []
        for line in fh:
            line = line.strip('\n')
            line = line.rstrip()
            words = line.split()
            imgs.append((words[0], int(words[1])))
        self.imgs = imgs
        self.transform = transform
        self.target_transform = target_transform
        self.loader = loader

    def __getitem__(self, index):
        fn, label = self.imgs[index]
        img = self.loader(fn)
        if self.transform is not None:
            img = self.transform(img)
        return img, label

    def __len__(self):
        return len(self.imgs)


train_data = MyDataset(txt=root + 'rawtrain.txt', transform=transforms.ToTensor())
test_data = MyDataset(txt=root + 'rawtest.txt', transform=transforms.ToTensor())
train_loader = DataLoader(dataset=train_data, batch_size=31, shuffle=True)
test_loader = DataLoader(dataset=test_data, batch_size=31, shuffle=True)
transform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.1307,), (0.3081,))
                                ])

3.模型训练设置

model = LeNet5()
#使用softmax分类
criterion = torch.nn.CrossEntropyLoss()
#设置随机梯度下降 学习率和L2正则
optimizer = optim.SGD(model.parameters(), lr=0.01, momentum=0.5)
#使用GPU训练
model = model.to(device)

4.训练

每训练一个epoch 做一次平均loss train acc test acc的计算绘制

def train(epoch):
    running_loss = 0.0
    num_correct = 0.0
    total = 0
    correct = 0
    total = 0
    test_acc = 0.0
    # train

    for batch_idx, data in enumerate(train_loader, 0):
        inputs, target = data
        inputs = inputs.to(device)
        target = target.to(device)
        optimizer.zero_grad()
        # forward + backward + update
        outputs = model(inputs)
        loss = criterion(outputs, target)

        loss.backward()
        optimizer.step()
        running_loss += loss.item()
        _, predicted = torch.max(outputs.data, dim=1)
        total += target.size(0)
        num_correct += (predicted == target).sum().item()
    # #test
    # with torch.no_grad():
    #     for data in test_loader:
    #         images, labels = data
    #         images = images.to(device)
    #         labels = labels.to(device)
    #         outputs = model(images)
    #         _, predicted = torch.max(outputs.data, dim=1)
    #         total += labels.size(0)
    #
    #         correct += (predicted == labels).sum().item()

    print('[%d, %5d] loss: %.3f' % (epoch + 1, batch_idx + 1, running_loss / len(train_loader)))
    # print('Accuracy on test set: %d %%' % (100 * correct / total))
    # test_acc=100 * correct / total
    test_acc = test()
    acc = (num_correct / len(train_loader.dataset) * 100)
    print("num_correct=")
    print(acc)
    running_loss /= len(train_loader)
    draw.new_data(running_loss, acc, test_acc, 2)
    draw.draw()

def test():
    correct = 0
    total = 0
    with torch.no_grad():
        for data in test_loader:
            images, labels = data
            images = images.to(device)
            labels = labels.to(device)
            outputs = model(images)
            _, predicted = torch.max(outputs.data, dim=1)
            total += labels.size(0)
            correct += (predicted == labels).sum().item()
    test_acc = 100 * correct / total
    print('Accuracy on test set: ', test_acc, '%')
    return test_acc

5.结果统计

if __name__ == '__main__':
    for epoch in range(20):
        train(epoch)

    torch.save(model.state_dict(), "minist_last.pth")
    draw.show()

从图中效果可以看到随着训练次数的增加，loss在不断下降，train acc 和test acc 也在慢慢收敛，最终达到了train acc=97% test acc=94%的效果，但发现曲线依然还有优化的空间，训练次数不是很足够。

这里我将训练参数保留了下来，然后接着在前面基础上继续训练，观察结果。

在继续训练的过程中出现了测试准确率突然下滑的现象，目前我不知道为什么，但最高的精度可达96.21%，效果还算不错。