论文复现--基于LeNet网络结构的数字识别

前言

• 一直就听说学习深度学习无非就是看论文，然后复现，不断循环，这段时间也看了好几篇论文(虽然都是简单的)，但是对于我一个人自学，复现成功，我感觉还是挺开心的

• 本人初学看论文的思路：聚焦网络结构与其实验的效果

• LeNet虽然简单，很老了，但是毕竟经典，对于初学的的我来说，我感觉还是很有必要学习的，可以积累CNN网络结构模型
• 注意：minist数据集可以直接下载，不用自己找，详情请看导入数据

• 本来今天打算更新C从C++的变化基础的，但是由于种种原因，就先更新这篇吧

论文(知网可查询)：基于LeNet-5的手写数字识别的改进方法

网络结构(LeNet)：

• 卷积层：两层

• 池化层：两层

• 卷积层参数：

  • 第一层：维度变化(1->6)，步伐：1，卷积核：5 * 5
  • 第二层：维度变化(6->16)，步伐：1，卷积核：5 * 5

• 池化层：

  • 两层都是：卷积核：2 * 2，步伐：2

• 全连接层：3层

  • 16 * 5 * 5 --> 120 --> 84 --> 10

• 网络结构图如下(论文截图)：

  

结果

• 轮次10，有点大了，可以降低
• 相比第一课，发现在训练集的损失率、测试集的损失率、训练集的准确率都有提升，详情情况结果可视化

1、前期准备

1、设置GPU

import torch  # 用于张量计算和自动求导
import torch.nn as nn  # 构建神经网络和损失函数
import matplotlib.pyplot as plt # 绘图
import torchvision   # 专门处理视觉的库# 设置GPU
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
print(device)
print(torch.__version__)
print(torchvision.__version__)

cuda
2.4.0
0.19.0

2、导入数据

# 将所有的数据图片统一格式, 论文大小为：32 * 32
from torchvision import transforms, datasets transforms = transforms.Compose([transforms.Resize([32, 32]),  # 统一图片大小transforms.ToTensor(),           # 统一规格transforms.Normalize(mean=[0.1307], std=[0.3081])  # MNIST的均值和方差
])

# download设置为True，可以自动下载图片
train_ds = torchvision.datasets.MNIST('data', train=True, transform=transforms, download=False)test_ds = torchvision.datasets.MNIST('data', train=True, transform=transforms, download=False)

batch_size = 32
train_dl = torch.utils.data.DataLoader(train_ds, batch_size=batch_size, shuffle=True)test_dl = torch.utils.data.DataLoader(test_ds, batch_size=batch_size, shuffle=True)

# 取一个批次查看数据格式
# 数据的shape为：[batch_size, channel, heigh, weight]
# batch_size是自己设定的，channel，height，weight分别是图片的通道数，高度，宽度
imgs, labels = next(iter(train_dl))
imgs.shape

结果：

torch.Size([32, 1, 32, 32])

3、数据可视化

import numpy as np# 指定图片的大小，图像的大小为20宽，5高
plt.figure(figsize=(20,5))
for i, imgs in enumerate(imgs[:20]):# 维度缩减npimg = np.squeeze(imgs.numpy())# 将整个figure分层2行10列，绘制第i+1个子图plt.subplot(2, 10, i + 1)plt.imshow(npimg, cmap=plt.cm.binary)plt.axis('off')

​

​

2、构建简单的CNN网络

import torch.nn.functional as Fnum_classes = 10 # 图片的类别数class Model(nn.Module):def __init__(self):super().__init__()# 特征提取网络设置self.conv1 = nn.Conv2d(1, 6, kernel_size=5) self.pool1 = nn.MaxPool2d(2)                 self.conv2 = nn.Conv2d(6, 16, kernel_size=5) self.pool2 = nn.MaxPool2d(2)                 # 分类网络设置self.fc1 = nn.Linear(16 * 5 * 5, 120)self.fc2 = nn.Linear(120, 84)self.fc3 = nn.Linear(84, num_classes)# 前向传播def forward(self, x):x = F.relu(self.conv1(x))x = self.pool1(x)x = F.relu(self.conv2(x))x = self.pool2(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        

加载并且打印模型

from torchinfo import summary# 将模型转移到GPU中
model = Model().to(device)summary(model)

=================================================================
Layer (type:depth-idx)                   Param #
=================================================================
Model                                    --
├─Conv2d: 1-1                            156
├─MaxPool2d: 1-2                         --
├─Conv2d: 1-3                            2,416
├─MaxPool2d: 1-4                         --
├─Linear: 1-5                            48,120
├─Linear: 1-6                            10,164
├─Linear: 1-7                            850
=================================================================
Total params: 61,706
Trainable params: 61,706
Non-trainable params: 0
=================================================================

for X, y in train_dl:print(X.shape)  # 检查输入数据的形状break  # 只打印第一个批次的数据形状

torch.Size([32, 1, 32, 32])

3、模型训练

1、设置超参数

loss_fn = nn.CrossEntropyLoss()  # 创建损失函数
learn_rate = 1e-2  # 学习率
opt = torch.optim.SGD(model.parameters(), lr = learn_rate)  

2、编写训练函数

def train(dataloader, model, loss_fn, optimizer):size = len(dataloader.dataset)      # 训练集大小一共60000张图片num_batchs = len(dataloader)        # 批次数目，1875  (60000/32)train_loss, train_acc = 0, 0  # 初始化训练损失和正确率for X, y in dataloader:  # 获取图片及其标签X, y = X.to(device), y.to(device)  # 计算预测误差pred = model(X)     # 网络输出loss = loss_fn(pred, y)     # 计算网络输出和真实值的差距# 反向传播optimizer.zero_grad()  # gred属性归零loss.backward()        # 反向传播optimizer.step()       # 每一步自动跟新# 记录acc和losstrain_acc += (pred.argmax(1) == y).type(torch.float).sum().item()train_loss += loss.item()train_acc /= sizetrain_loss /= num_batchsreturn train_acc, train_loss

3、编写测试函数

def test(dataloader, model, loss_fn):size = len(dataloader.dataset)  # 测试集的大小，一共10000张图片num_batches = len(dataloader)   # 批次数目，313（10000/32 = 321.5，向上取整）test_loss, test_acc = 0, 0# 当不进行训练时候，停止梯度更新，节省计算内存消耗with torch.no_grad():for imgs, target in dataloader:imgs, target = imgs.to(device), target.to(device)# 计算losstarget_pred = model(imgs)loss = loss_fn(target_pred, target)test_loss += loss.item()test_acc += (target_pred.argmax(1) == target).type(torch.float).sum().item()test_acc /= sizetest_loss /= num_batchesreturn test_acc, test_loss

4、正式训练

epochs = 10
train_loss = []
train_acc = []
test_loss = []
test_acc = []for epoch in range(epochs):model.train()epoch_train_acc, epoch_train_loss = train(train_dl, model, loss_fn, opt)model.eval()epoch_test_acc, epoch_test_loss = test(test_dl, model, loss_fn)train_acc.append(epoch_train_acc)train_loss.append(epoch_train_loss)test_acc.append(epoch_test_acc)test_loss.append(epoch_test_loss)template = ('Eopch: {:2d}, Train_acc: {:.1f}%, Train_loss: {:.3f}, Test_acc: {:.1f}%, test_loss: {:.3f}')print(template.format(epoch+1, epoch_train_acc * 100, epoch_train_loss, epoch_test_acc, epoch_test_loss))print('Done')

Eopch:  1, Train_acc: 75.9%, Train_loss: 0.739, Test_acc: 1.0%, test_loss: 0.144
Eopch:  2, Train_acc: 96.4%, Train_loss: 0.117, Test_acc: 1.0%, test_loss: 0.079
Eopch:  3, Train_acc: 97.6%, Train_loss: 0.080, Test_acc: 1.0%, test_loss: 0.073
Eopch:  4, Train_acc: 98.0%, Train_loss: 0.063, Test_acc: 1.0%, test_loss: 0.056
Eopch:  5, Train_acc: 98.4%, Train_loss: 0.053, Test_acc: 1.0%, test_loss: 0.048
Eopch:  6, Train_acc: 98.5%, Train_loss: 0.047, Test_acc: 1.0%, test_loss: 0.041
Eopch:  7, Train_acc: 98.7%, Train_loss: 0.042, Test_acc: 1.0%, test_loss: 0.035
Eopch:  8, Train_acc: 98.8%, Train_loss: 0.037, Test_acc: 1.0%, test_loss: 0.029
Eopch:  9, Train_acc: 99.0%, Train_loss: 0.033, Test_acc: 1.0%, test_loss: 0.029
Eopch: 10, Train_acc: 99.0%, Train_loss: 0.030, Test_acc: 1.0%, test_loss: 0.023
Done

4、结果可视化

import matplotlib.pyplot as plt
import warnings
# 忽略警告
warnings.filterwarnings("ignore")               #忽略警告信息
plt.rcParams['font.sans-serif']    = ['SimHei'] # 用来正常显示中文标签
plt.rcParams['axes.unicode_minus'] = False      # 用来正常显示负号
plt.rcParams['figure.dpi']         = 100        #分辨率epochs_range = range(epochs)plt.figure(figsize=(12, 3))
plt.subplot(1, 2, 1)plt.plot(epochs_range, train_acc, label='Training Accuracy')
plt.plot(epochs_range, test_acc, label='Test Accuracy')
plt.legend(loc='lower right')
plt.title('Training and Validation Accuracy')plt.subplot(1, 2, 2)
plt.plot(epochs_range, train_loss, label='Train Loss')
plt.plot(epochs_range, test_loss, label='Test Loss')
plt.legend(loc='upper right')
plt.title('Training and Validation Loss')
plt.show()

​

​