深度学习 Day26——使用Pytorch实现猴痘病识别

一、前言

🍨 本文为🔗365天深度学习训练营 中的学习记录博客

🍦 参考文章：Pytorch实战 | 第P4周：猴痘病识别

🍖 原作者：K同学啊|接辅导、项目定制

这期博客在之前的猴痘病识别的基础上添加了指定图片预测与保存并加载模型这两个模块，将来我们训练后的模型是需要部署到真实环境中去测试的。

二、我的环境

• 电脑系统：Windows 11
• 语言环境：Python 3.8.5
• 编译器：Datalore
• 深度学习环境：
  • torch 1.12.1+cu113
  • torchvision 0.13.1+cu113
• 显卡及显存：RTX 3070 8G

三、前期工作

1、设置GPU导入依赖项

如果设备支持GPU就使用GPU，否则就是用CPU，但推荐深度学习使用GPU，如果设备不行，可以去网上云平台跑模型。

import torch
import torch.nn as nn
import torchvision.transforms as transforms
import torchvision
from torchvision import transforms, datasets
import os,PIL,pathlib

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

device

device(type='cuda')

2、导入猴痘病数据集

import os,PIL,random,pathlib

data_dir = 'E:\\深度学习\\data\\Day13'
data_dir = pathlib.Path(data_dir)

data_paths = list(data_dir.glob('*'))
classeNames = [str(path).split("\\")[4] for path in data_paths]
classeNames

['Monkeypox', 'Others']

total_datadir = 'E:\\深度学习\\data\\Day13'

train_transforms = transforms.Compose([
    transforms.Resize([224, 224]),  # 将输入图片resize成统一尺寸
    transforms.ToTensor(),          # 将PIL Image或numpy.ndarray转换为tensor，并归一化到[0,1]之间
    transforms.Normalize(           # 标准化处理-->转换为标准正太分布（高斯分布），使模型更容易收敛
        mean=[0.485, 0.456, 0.406],
        std=[0.229, 0.224, 0.225])  # 其中 mean=[0.485,0.456,0.406]与std=[0.229,0.224,0.225] 从数据集中随机抽样计算得到的。
])

total_data = datasets.ImageFolder(total_datadir,transform=train_transforms)
total_data

Dataset ImageFolder
    Number of datapoints: 2142
    Root location: E:\深度学习\data\Day13
    StandardTransform
Transform: Compose(
               Resize(size=[224, 224], interpolation=bilinear, max_size=None, antialias=None)
               ToTensor()
               Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
           )

total_data.class_to_idx

{'Monkeypox': 0, 'Others': 1}

3、划分数据集

将总数据集分为训练集和测试集，其中训练集占总数据集的80%，测试集占20%

train_size = int(0.8 * len(total_data))
test_size = len(total_data) - train_size
train_dataset, test_dataset = torch.utils.data.random_split(total_data, [train_size, test_size])
train_size, test_size

(1713, 429)

将训练集和测试集分别封装成 DataLoader 对象，方便对数据进行批量处理，batch_size 表示每个 batch 的大小，shuffle 表示是否随机打乱数据，num_workers 表示使用多少个线程来读取数据

train_loader = torch.utils.data.DataLoader(train_dataset, batch_size=32, shuffle=True, num_workers=1)
test_loader = torch.utils.data.DataLoader(test_dataset, batch_size=32, shuffle=True, num_workers=1)

for X, y in test_loader:
    print(X.shape, y.shape)
    break

torch.Size([32, 3, 224, 224]) torch.Size([32])

四、构建CNN网络

接下来我们定义一个简单的CNN网络结构。

import torch.nn.functional as F

# 定义一个带有Batch Normalization的卷积神经网络
class Network_bn(nn.Module):
    def __init__(self):
        super(Network_bn, self).__init__()
        """
        nn.Conv2d()函数：
        第一个参数（in_channels）是输入的channel数量
        第二个参数（out_channels）是输出的channel数量
        第三个参数（kernel_size）是卷积核大小
        第四个参数（stride）是步长，默认为1
        第五个参数（padding）是填充大小，默认为0
        """
        # 第一个卷积层，输入的channel数量是3，输出的channel数量是12，卷积核大小为5，步长为1，填充大小为0
        self.conv1 = nn.Conv2d(in_channels=3, out_channels=12, kernel_size=5, stride=1, padding=0)
        self.bn1 = nn.BatchNorm2d(12) # Batch Normalization层，输入的channel数量是12
        # 第二个卷积层，输入的channel数量是12，输出的channel数量是12，卷积核大小为5，步长为1，填充大小为0
        self.conv2 = nn.Conv2d(in_channels=12, out_channels=12, kernel_size=5, stride=1, padding=0)
        self.bn2 = nn.BatchNorm2d(12) # Batch Normalization层，输入的channel数量是12
        # 最大池化层，池化核大小为2，步长为2
        self.pool = nn.MaxPool2d(2,2)
        # 第三个卷积层，输入的channel数量是12，输出的channel数量是24，卷积核大小为5，步长为1，填充大小为0
        self.conv4 = nn.Conv2d(in_channels=12, out_channels=24, kernel_size=5, stride=1, padding=0)
        self.bn4 = nn.BatchNorm2d(24) # Batch Normalization层，输入的channel数量是24
        # 第四个卷积层，输入的channel数量是24，输出的channel数量是24，卷积核大小为5，步长为1，填充大小为0
        self.conv5 = nn.Conv2d(in_channels=24, out_channels=24, kernel_size=5, stride=1, padding=0)
        self.bn5 = nn.BatchNorm2d(24) # Batch Normalization层，输入的channel数量是24
        # 全连接层，输入的大小是24*50*50，输出的大小是类别数
        self.fc1 = nn.Linear(24*50*50, len(classeNames))

    # 定义网络的前向传播过程
    def forward(self, x):
        x = F.relu(self.bn1(self.conv1(x))) # 第一层卷积、Batch Normalization和ReLU激活函数
        x = F.relu(self.bn2(self.conv2(x))) # 第二层卷积、Batch Normalization和ReLU激活函数
        x = self.pool(x) # 最大池化层
        x = F.relu(self.bn4(self.conv4(x))) 
        x = F.relu(self.bn5(self.conv5(x))) # 第五层卷积、Batch Normalization和ReLU激活函数
        x = self.pool(x) # 最大池化层
        x = x.view(-1, 24*50*50) # 将卷积层的输出展平成一维向量
        x = self.fc1(x) # 全连接层
        return x

device = "cuda" if torch.cuda.is_available() else "cpu"
print("Using {} device".format(device))

model = Network_bn().to(device)
model

Using cuda device
Network_bn(
  (conv1): Conv2d(3, 12, kernel_size=(5, 5), stride=(1, 1))
  (bn1): BatchNorm2d(12, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (conv2): Conv2d(12, 12, kernel_size=(5, 5), stride=(1, 1))
  (bn2): BatchNorm2d(12, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (pool): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  (conv4): Conv2d(12, 24, kernel_size=(5, 5), stride=(1, 1))
  (bn4): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (conv5): Conv2d(24, 24, kernel_size=(5, 5), stride=(1, 1))
  (bn5): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
  (fc1): Linear(in_features=60000, out_features=2, bias=True)
)

五、训练模型

1、设置超参数

loss_fn = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(model.parameters(), lr=1e-3)

2、编写训练函数

我们自己定义一个训练函数train，该函数接受四个参数：dataloader，model，loss_fn和optimizer。其中，dataloader是一个PyTorch的数据加载器，用于加载训练数据；model是一个PyTorch的神经网络模型；loss_fn是一个损失函数，用于计算模型的预测误差；optimizer是一个优化器，用于更新模型的参数。函数的返回值是训练误差和训练精度。

在函数中，首先初始化训练误差和训练精度为0，然后遍历训练数据集中的每个批次。对于每个批次，首先将输入数据和标签数据转换为指定的设备（如GPU）上的张量，然后将输入数据输入模型，得到模型的预测结果。接着，使用损失函数计算模型的预测误差，并根据误差进行反向传播和参数更新。最后，累计训练误差和训练精度，并在每训练100个批次时输出当前的训练误差。最后，计算训练误差和训练精度的平均值，并输出训练误差和训练精度。该函数的作用是完成深度学习模型的训练过程，将输入数据经过模型计算得到输出结果，并根据损失函数计算输出结果与标签之间的差异，从而优化模型的参数，使模型能够更好地拟合数据。

def train(dataloader, model, loss_fn, optimizer):
    size = len(dataloader.dataset)
    num_batches = len(dataloader)
    train_loss, train_acc = 0, 0
    for batch, (X, y) in enumerate(dataloader):
        X, y = X.to(device), y.to(device)

        # 计算预测误差
        pred = model(X)
        loss = loss_fn(pred, y)

        # 反向传播
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()

        train_loss += loss.item()
        train_acc += (pred.argmax(1) == y).type(torch.float).sum().item()

        if batch % 100 == 0:
            loss, current = loss.item(), batch * len(X)
            print(f"loss: {loss:>7f}  [{current:>5d}/{size:>5d}]")
    train_loss /= num_batches
    train_acc /= size
    print(f"Train Error: \n Accuracy: {(100*train_acc):>0.1f}%, Avg loss: {train_loss:>8f} \n")
    return train_loss, train_acc

3、编写测试函数

测试函数和训练函数大致相同，但是由于不进行梯度下降对网络权重进行更新，所以不需要传入优化器。

我们自己定义一个训练函数test函数，模型在测试集上的评估，包括计算测试集上的损失和准确率。

具体来说，test函数接受一个数据集迭代器、一个模型、一个损失函数作为输入，并返回模型在测试集上的平均损失和准确率。

函数的具体实现如下：

• 首先获取数据集大小和批次数量，并将模型设为评估模式；
• 然后遍历测试集迭代器，将每个batch的数据送入模型计算预测值；
• 用预测值和实际标签计算损失，并将损失值和准确率累加到test_loss和test_acc变量中；
• 最后除以批次数得到平均损失和准确率，并打印输出结果。

需要注意的是，由于在测试集上不需要反向传播计算梯度，因此需要使用torch.no_grad()上下文管理器来禁用梯度计算，从而提高计算效率。

def test(dataloader, model, loss_fn):
    size = len(dataloader.dataset)
    num_batches = len(dataloader)
    model.eval()
    test_loss, test_acc = 0, 0
    with torch.no_grad():
        for X, y in dataloader:
            X, y = X.to(device), y.to(device)
            pred = model(X)
            test_loss += loss_fn(pred, y).item()
            test_acc += (pred.argmax(1) == y).type(torch.float).sum().item()
    test_loss /= num_batches
    test_acc /= size
    print(f"Test Error: \n Accuracy: {(100*test_acc):>0.1f}%, Avg loss: {test_loss:>8f} \n")
    return  test_loss, test_acc

4、正式训练

epochs = 20
train_loss, train_acc = [], []
test_loss, test_acc = [], []
for t in range(epochs):
    model.train()
    epoch_train_acc, epoch_train_loss = train(train_loader, model, loss_fn, optimizer)
    train_loss.append(epoch_train_loss)
    train_acc.append(epoch_train_acc)
    model.eval()
    epoch_test_acc, epoch_test_loss = test(test_loader, model, loss_fn)
    test_loss.append(epoch_test_loss)
    test_acc.append(epoch_test_acc)
    print(f"Epoch {t+1}\n-------------------------------")
    train(train_loader, model, loss_fn, optimizer)
    test(test_loader, model, loss_fn)
print("Done!")

训练结果为：

loss: 1.199974  [    0/ 1713]
Train Error: 
 Accuracy: 83.9%, Avg loss: 0.503470 

Test Error: 
 Accuracy: 86.2%, Avg loss: 0.487529 

Epoch 1
-------------------------------
loss: 0.062999  [    0/ 1713]
Train Error: 
 Accuracy: 81.9%, Avg loss: 0.463360 

Test Error: 
 Accuracy: 79.7%, Avg loss: 0.457709 

loss: 0.708710  [    0/ 1713]
Train Error: 
 Accuracy: 85.9%, Avg loss: 0.406817 

Test Error: 
 Accuracy: 85.3%, Avg loss: 0.506047 

Epoch 2
-------------------------------
loss: 0.270929  [    0/ 1713]
Train Error: 
 Accuracy: 87.2%, Avg loss: 0.338913 

Test Error: 
 Accuracy: 82.8%, Avg loss: 0.500975 

loss: 0.657500  [    0/ 1713]
Train Error: 
 Accuracy: 87.1%, Avg loss: 0.405702 

Test Error: 
 Accuracy: 80.4%, Avg loss: 0.691268 

Epoch 3
-------------------------------
loss: 0.149799  [    0/ 1713]
Train Error: 
 Accuracy: 87.9%, Avg loss: 0.310380 

Test Error: 
 Accuracy: 79.7%, Avg loss: 0.514998 

loss: 0.230001  [    0/ 1713]
Train Error: 
 Accuracy: 85.9%, Avg loss: 0.478800 

Test Error: 
 Accuracy: 76.9%, Avg loss: 1.450183 

Epoch 4
-------------------------------
loss: 0.731624  [    0/ 1713]
Train Error: 
 Accuracy: 84.7%, Avg loss: 0.388541 

Test Error: 
 Accuracy: 83.2%, Avg loss: 0.551030 

loss: 0.425535  [    0/ 1713]
Train Error: 
 Accuracy: 87.9%, Avg loss: 0.339442 

Test Error: 
 Accuracy: 84.6%, Avg loss: 0.762711 

Epoch 5
-------------------------------
loss: 0.261778  [    0/ 1713]
Train Error: 
 Accuracy: 91.4%, Avg loss: 0.251970 

Test Error: 
 Accuracy: 84.1%, Avg loss: 0.550993 

loss: 0.120489  [    0/ 1713]
Train Error: 
 Accuracy: 92.3%, Avg loss: 0.267334 

Test Error: 
 Accuracy: 82.1%, Avg loss: 0.732856 

Epoch 6
-------------------------------
loss: 0.545078  [    0/ 1713]
Train Error: 
 Accuracy: 93.5%, Avg loss: 0.190953 

Test Error: 
 Accuracy: 81.4%, Avg loss: 0.654517 

loss: 0.242050  [    0/ 1713]
Train Error: 
 Accuracy: 93.9%, Avg loss: 0.166487 

Test Error: 
 Accuracy: 86.5%, Avg loss: 0.520211 

Epoch 7
-------------------------------
loss: 0.032337  [    0/ 1713]
Train Error: 
 Accuracy: 94.4%, Avg loss: 0.139460 

Test Error: 
 Accuracy: 81.8%, Avg loss: 0.571910 

loss: 0.111919  [    0/ 1713]
Train Error: 
 Accuracy: 95.3%, Avg loss: 0.133110 

Test Error: 
 Accuracy: 87.9%, Avg loss: 0.581790 

Epoch 8
-------------------------------
loss: 0.011513  [    0/ 1713]
Train Error: 
 Accuracy: 93.2%, Avg loss: 0.168966 

Test Error: 
 Accuracy: 84.6%, Avg loss: 0.612067 

loss: 0.138732  [    0/ 1713]
Train Error: 
 Accuracy: 95.9%, Avg loss: 0.131339 

Test Error: 
 Accuracy: 85.5%, Avg loss: 0.658192 

Epoch 9
-------------------------------
loss: 0.098304  [    0/ 1713]
Train Error: 
 Accuracy: 97.5%, Avg loss: 0.073731 

Test Error: 
 Accuracy: 86.2%, Avg loss: 0.648000 

loss: 0.041354  [    0/ 1713]
Train Error: 
 Accuracy: 98.2%, Avg loss: 0.054499 

Test Error: 
 Accuracy: 85.1%, Avg loss: 0.804457 

Epoch 10
-------------------------------
loss: 0.111462  [    0/ 1713]
Train Error: 
 Accuracy: 97.4%, Avg loss: 0.069474 

Test Error: 
 Accuracy: 86.9%, Avg loss: 0.575027 

loss: 0.039980  [    0/ 1713]
Train Error: 
 Accuracy: 97.8%, Avg loss: 0.069207 

Test Error: 
 Accuracy: 86.5%, Avg loss: 0.715076 

Epoch 11
-------------------------------
loss: 0.030235  [    0/ 1713]
Train Error: 
 Accuracy: 97.5%, Avg loss: 0.078401 

Test Error: 
 Accuracy: 87.2%, Avg loss: 0.594295 

loss: 0.055335  [    0/ 1713]
Train Error: 
 Accuracy: 97.5%, Avg loss: 0.059249 

Test Error: 
 Accuracy: 87.4%, Avg loss: 0.696577 

Epoch 12
-------------------------------
loss: 0.040502  [    0/ 1713]
Train Error: 
 Accuracy: 96.5%, Avg loss: 0.100692 

Test Error: 
 Accuracy: 82.1%, Avg loss: 0.793313 

loss: 0.044457  [    0/ 1713]
Train Error: 
 Accuracy: 95.9%, Avg loss: 0.108039 

Test Error: 
 Accuracy: 84.6%, Avg loss: 0.848924 

Epoch 13
-------------------------------
loss: 0.022895  [    0/ 1713]
Train Error: 
 Accuracy: 98.2%, Avg loss: 0.055400 

Test Error: 
 Accuracy: 86.2%, Avg loss: 0.881824 

loss: 0.058409  [    0/ 1713]
Train Error: 
 Accuracy: 97.5%, Avg loss: 0.066015 

Test Error: 
 Accuracy: 86.5%, Avg loss: 0.834848 

Epoch 14
-------------------------------
loss: 0.037372  [    0/ 1713]
Train Error: 
 Accuracy: 99.2%, Avg loss: 0.029454 

Test Error: 
 Accuracy: 86.5%, Avg loss: 0.886635 

loss: 0.126379  [    0/ 1713]
Train Error: 
 Accuracy: 98.9%, Avg loss: 0.036465 

Test Error: 
 Accuracy: 86.7%, Avg loss: 0.926361 

Epoch 15
-------------------------------
loss: 0.022206  [    0/ 1713]
Train Error: 
 Accuracy: 96.1%, Avg loss: 0.134624 

Test Error: 
 Accuracy: 83.0%, Avg loss: 0.761580 

loss: 0.109468  [    0/ 1713]
Train Error: 
 Accuracy: 95.0%, Avg loss: 0.145105 

Test Error: 
 Accuracy: 86.2%, Avg loss: 0.864981 

Epoch 16
-------------------------------
loss: 0.116144  [    0/ 1713]
Train Error: 
 Accuracy: 97.8%, Avg loss: 0.056436 

Test Error: 
 Accuracy: 85.5%, Avg loss: 0.808745 

loss: 0.148035  [    0/ 1713]
Train Error: 
 Accuracy: 98.2%, Avg loss: 0.056249 

Test Error: 
 Accuracy: 87.2%, Avg loss: 0.805620 

Epoch 17
-------------------------------
loss: 0.107752  [    0/ 1713]
Train Error: 
 Accuracy: 98.9%, Avg loss: 0.028704 

Test Error: 
 Accuracy: 85.3%, Avg loss: 0.989487 

loss: 0.005748  [    0/ 1713]
Train Error: 
 Accuracy: 99.2%, Avg loss: 0.027402 

Test Error: 
 Accuracy: 86.0%, Avg loss: 0.791777 

Epoch 18
-------------------------------
loss: 0.005322  [    0/ 1713]
Train Error: 
 Accuracy: 99.5%, Avg loss: 0.015000 

Test Error: 
 Accuracy: 86.2%, Avg loss: 0.807837 

loss: 0.003800  [    0/ 1713]
Train Error: 
 Accuracy: 99.4%, Avg loss: 0.020819 

Test Error: 
 Accuracy: 84.4%, Avg loss: 1.052223 

Epoch 19
-------------------------------
loss: 0.001303  [    0/ 1713]
Train Error: 
 Accuracy: 99.4%, Avg loss: 0.015747 

Test Error: 
 Accuracy: 85.8%, Avg loss: 1.024608 

loss: 0.007683  [    0/ 1713]
Train Error: 
 Accuracy: 98.5%, Avg loss: 0.067711 

Test Error: 
 Accuracy: 85.3%, Avg loss: 1.076210 

Epoch 20
-------------------------------
loss: 0.001310  [    0/ 1713]
Train Error: 
 Accuracy: 94.2%, Avg loss: 0.196802 

Test Error: 
 Accuracy: 84.8%, Avg loss: 0.813763 

Done!

六、结果可视化

# 可视化上述训练结果
import matplotlib.pyplot as plt

def plot_curve(train_loss, val_loss, train_acc, val_acc):
    plt.figure(figsize=(8, 8))
    plt.subplot(2, 1, 1)
    plt.plot(train_loss, label='train loss')
    plt.plot(val_loss, label='val loss')
    plt.legend(loc='best')
    plt.xlabel('Epochs')
    plt.ylabel('Loss')
    plt.subplot(2, 1, 2)
    plt.plot(train_acc, label='train acc')
    plt.plot(val_acc, label='val acc')
    plt.legend(loc='best')
    plt.xlabel('Epochs')
    plt.ylabel('Accuracy')
    plt.show()
plot_curve(train_loss, test_loss, train_acc, test_acc)

七、图片预测

from PIL import Image
classes = list(total_data.class_to_idx)

def predict_one_image(image_path, model, transform, classes):

    test_img = Image.open(image_path).convert('RGB')
    # plt.imshow(test_img)  # 展示预测的图片

    test_img = transform(test_img)
    img = test_img.to(device).unsqueeze(0)

    model.eval()
    output = model(img)

    _,pred = torch.max(output,1)
    pred_class = classes[pred]
    print(f'预测结果是：{pred_class}')
predict_one_image('E:\\深度学习\\data\\Day13\\Monkeypox\\M01_01_00.jpg', model, train_transforms, classes)

预测结果是：Monkeypox

预测结果是正确的，但是准确率没有到88%。

八、保存模型

# 模型保存
torch.save(model.state_dict(), "model.pth")

# 将参数加载到model当中
model.load_state_dict(torch.load(PATH, map_location=device))
print("Saved PyTorch Model State to model.pth")

Saved PyTorch Model State to model.pth

九、模型优化

我添加了一层Dropout层，但是最后训练的准确率并没有提升。

...
Epoch 20
-------------------------------
loss: 0.019593  [    0/ 1713]
Train Error: 
 Accuracy: 99.1%, Avg loss: 0.023408 

Test Error: 
 Accuracy: 84.6%, Avg loss: 0.888081