从入门到深入卷积神经网络学习分享（关于对Pytorch的MNIST手写字母的识别）

笔者采用4种网络层次结构来实现关于MNIST手写字母识别的神经网络，从表层开始步步深入，便于理解。

四种网络分别是：

一：3层全连接网络模型（包含批处理化以及ReLU优化函数）

  (layer1): Sequential(
    (0): Linear(in_features=784, out_features=300, bias=True)
    (1): BatchNorm1d(300, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU(inplace=True)
  )
  (layer2): Sequential(
    (0): Linear(in_features=300, out_features=100, bias=True)
    (1): BatchNorm1d(100, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU(inplace=True)
  )
  (layer3): Sequential(
    (0): Linear(in_features=100, out_features=10, bias=True)
  )

二：2层卷积神经网络和3层全连接输出（无批处理化有ReLU优化函数）

（此2层卷积转载来自其他作者，如侵必删）

        self.conv1 = nn.Sequential(     #input_size=(1*28*28)
            nn.Conv2d(1, 6, 5, 1, 2), #padding=2保证输入输出尺寸相同
            nn.ReLU(),      #input_size=(6*28*28)
            nn.MaxPool2d(kernel_size=2, stride=2),#output_size=(6*14*14)
        )
        self.conv2 = nn.Sequential(
            nn.Conv2d(6, 16, 5),
            nn.ReLU(),      #input_size=(16*10*10)
            nn.MaxPool2d(2, 2)  #output_size=(16*5*5)
        )
        self.fc1 = nn.Sequential(
            nn.Linear(16 * 5 * 5, 120),
            nn.ReLU()
        )
        self.fc2 = nn.Sequential(
            nn.Linear(120, 84),
            nn.ReLU()
        )
        self.fc3 = nn.Linear(84, 10)

三：4层卷积神经网络和3层全连接输出（无批处理化有ReLU函数优化）

 

CNN(
  (layer1): Sequential(
    (0): Conv2d(1, 16, kernel_size=(3, 3), stride=(1, 1))
    (1): ReLU(inplace=True)
  )
  (layer2): Sequential(
    (0): Conv2d(16, 32, kernel_size=(3, 3), stride=(1, 1))
    (1): ReLU(inplace=True)
    (2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (layer3): Sequential(
    (0): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1))
    (1): ReLU(inplace=True)
  )
  (layer4): Sequential(
    (0): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1))
    (1): ReLU(inplace=True)
    (2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (fc): Sequential(
    (0): Linear(in_features=2048, out_features=1024, bias=True)
    (1): ReLU(inplace=True)
    (2): Linear(in_features=1024, out_features=128, bias=True)
    (3): ReLU(inplace=True)
    (4): Linear(in_features=128, out_features=10, bias=True)
  )
)

测试集正确率明显有很大提升，函数收敛速度加快几倍，笔者会在最后附上正确率对比

四：4层卷积神经网络和3层全连接输出（有批处理化有ReLU函数优化）—最优

  (layer1): Sequential(
    (0): Conv2d(1, 16, kernel_size=(3, 3), stride=(1, 1))
    (1): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU(inplace=True)
  )
  (layer2): Sequential(
    (0): Conv2d(16, 32, kernel_size=(3, 3), stride=(1, 1))
    (1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU(inplace=True)
    (3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (layer3): Sequential(
    (0): Conv2d(32, 64, kernel_size=(3, 3), stride=(1, 1))
    (1): BatchNorm2d(64, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU(inplace=True)
  )
  (layer4): Sequential(
    (0): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1))
    (1): BatchNorm2d(128, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU(inplace=True)
    (3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (fc): Sequential(
    (0): Linear(in_features=2048, out_features=1024, bias=True)
    (1): ReLU(inplace=True)
    (2): Linear(in_features=1024, out_features=128, bias=True)
    (3): ReLU(inplace=True)
    (4): Linear(in_features=128, out_features=10, bias=True)
  )

上述第4种模型采用卷积层-批处理化-ReLU优化-池化层 这种顺序搭建

第一种3层全连接模型采用20次训练结果，其他模型采用8次训练，因为第一种模型如果采用8次训练，结果并未收敛无参考价值

最后笔者列出第第四种4层卷积包含批处理化模型全部代码，供参考学习

CNN2_Net.py

import torch
from torch import nn, optim
from torch.autograd import Variable
from torch.utils.data import DataLoader
from torchvision import datasets, transforms


class CNN(nn.Module):
    def __init__(self):
        super(CNN, self).__init__()
        self.layer1 = nn.Sequential(
            nn.Conv2d(1, 16, kernel_size=3), nn.BatchNorm2d(16), nn.ReLU(True))

        self.layer2 = nn.Sequential(
            nn.Conv2d(16, 32, kernel_size=3), nn.BatchNorm2d(32), nn.ReLU(True), nn.MaxPool2d(2, 2))

        self.layer3 = nn.Sequential(
            nn.Conv2d(32, 64, kernel_size=3), nn.BatchNorm2d(64), nn.ReLU(True))

        self.layer4 = nn.Sequential(
            nn.Conv2d(64, 128, kernel_size=3), nn.BatchNorm2d(128), nn.ReLU(True), nn.MaxPool2d(2, 2))

        self.fc = nn.Sequential(
            nn.Linear(128 * 4 * 4, 1024), nn.ReLU(inplace=True), nn.Linear(1024, 128), nn.ReLU(inplace=True),
            nn.Linear(128, 10))

    def forward(self, x):
        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)
        x = x.view(x.size(0), -1)
        x = self.fc(x)
        return x

main.py

import torch
import torchvision as tv
import torchvision.transforms as transforms
import torch.nn as nn
import torch.optim as optim
import CNN2_Net

# 超参数设置
EPOCH = 8  # 遍历数据集次数
BATCH_SIZE = 64  # 批处理尺寸(batch_size)
LR = 0.001  # 学习率

# 定义数据预处理方式
transform = transforms.ToTensor()

# 定义训练数据集
trainset = tv.datasets.MNIST(
    root='./data/',
    train=True,
    download=True,
    transform=transform)

# 定义训练批处理数据
trainloader = torch.utils.data.DataLoader(
    trainset,
    batch_size=BATCH_SIZE,
    shuffle=True,
)

# 定义测试数据集
testset = tv.datasets.MNIST(
    root='./data/',
    train=False,
    download=True,
    transform=transform)

# 定义测试批处理数据
testloader = torch.utils.data.DataLoader(
    testset,
    batch_size=BATCH_SIZE,
    shuffle=False,
)

# 定义损失函数loss function 和优化方式（采用SGD）
net = CNN2_Net.CNN()
criterion = nn.CrossEntropyLoss()  # 交叉熵损失函数，通常用于多分类问题上
optimizer = optim.SGD(net.parameters(), lr=LR, momentum=0.9)

# 训练
if __name__ == "__main__":

    for epoch in range(EPOCH):
        sum_loss = 0.0
        Total_sum_loss = 0.0
        # 数据读取
        for i, data in enumerate(trainloader):
            inputs, labels = data
            # inputs, labels = inputs.to(device), labels.to(device)

            # 梯度清零
            optimizer.zero_grad()

            # forward + backward
            outputs = net(inputs)
            loss = criterion(outputs, labels)
            loss.backward()
            optimizer.step()
            # 每训练100个batch打印一次平均loss
            sum_loss += loss.item()
            Total_sum_loss += loss.item()
            if i % 100 == 99:
                print('[%d, %d] loss: %.03f'
                      % (epoch + 1, i + 1, sum_loss / 100))
                sum_loss = 0.0
        print('[%d] Total_loss: %.03f' % (epoch + 1, Total_sum_loss / len(trainloader)))
        # 每跑完一次epoch测试一下准确率
        with torch.no_grad():
            correct = 0
            total = 0
            i = 0
            for data in testloader:
                images, labels = data
                # images, labels = images.to(device), labels.to(device)
                outputs = net(images)
                # 取得分最高的那个类
                _, predicted = torch.max(outputs.data, 1)  # 选择概率最大的那个类别
                total += labels.size(0)
                i += 1
                correct += (predicted == labels).sum()
                print(labels.size(0))
            print('第%d个epoch的识别准确率为：%d%%' % (epoch + 1, (100 * correct / total)))

大家可以在下方留言