526互联

【深度学习】PyTorch 图片识别分类器

发布时间 2023-11-02 14:40:24作者: PythonNew_Mr.Wang

下载数据集

import torch
import torchvision
import torchvision.transforms as transforms

# 定义数据预处理的转换
transform = transforms.Compose(
    [transforms.ToTensor(),
     transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])

# 训练集
trainset = torchvision.datasets.CIFAR10(root='./data', train=True,download=True, transform=transform)

# 加载
trainloader = torch.utils.data.DataLoader(trainset, batch_size=4,shuffle=True, num_workers=2)

# 测试集
testset = torchvision.datasets.CIFAR10(root='./data', train=False,download=True, transform=transform)

# 加载
testloader = torch.utils.data.DataLoader(testset, batch_size=4,shuffle=False, num_workers=2)

构建神经网络

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

class Net(nn.Module):
  
  def __init__(self):
    super(Net,self).__init__()
    
    #  定义第一层卷积层
    self.c1 = nn.Conv2d(3 ,6 ,5)
    self.c2 = nn.Conv2d(6 ,15, 5)
    
    # 定义池化
    self.pool = nn.MaxPool2d(2,2)
    
    # 定义全连接
    self.fc1 = nn.Linear(15 * 5 * 5,120)
    self.fc2 = nn.Linear(120 , 84)
    slef.fc3 = nn.Linear(84 , 10)
    
  def forward(self):
    
    # 卷积层  relu激活函数去负值  池化
    x = self.pool(F.relu(self.c1(x)))
    x = self.pool(F.relu(self.c2(x)))
    
    # 扁平化  转换成x形状 适配 全连接层的输入
    x = x.view(-1, 15 * 5 * 5)
    
    # 全连接层 - relu激活函数 去负值  
    x = F.relu(self.fc1(x))
    x = F.relu(self.fc2(x))
    x = F.slef.fc3(x)
    
   	return x
    
net = Net()

训练模型

# 损失函数 与 梯度清零函数
import torch.optim as optim
losser = nn.CrossEntropyLoss()
optimer = optim.SGD(net.parameters(),lr=0.1)

running_loss = 0

# 训练次数
for epoch in range(2):
  # 遍历训练集
  for data in trainloader:
    
    # 提取图片与标签
    images, labels = data
    
    # 梯度清理
    optimer.zero_grad()
    
    # 计算训练结果张量
    outputs = net(images)
    # 计算损失值
    loss = losser(outputs,labels)
   	# 反向传播
    loss.backward()
    
    # 参数更新
    optimer.step()
   	
    # 打印训练数据
    running_loss += loss.item()
    if (i + 1) % 2000 == 0:
      print('[{}, {}] loss: {}'.format(epoch + 1, i + 1, running_loss / 2000))
      running_loss = 0
     
print("训练完毕")

保存模型

torch.save(net.state_dict(),'data.pth')

预测模型

correct = 0  #  正确率
total = 0    #  样本总量

# 禁止梯度清零  （测试/开发）
with torch.no_grad():
  # 遍历测试集
  for data in testloader:
    
    images,labels = data
    outputs = net(images)
    _,predicted = torch.max(outputs,1) # 预测
    
    total += lables.size(0) # 总量
    
    # sum()方法用于计算张量中所有元素的总和
    # item()将张量值转换成标量值
    correct += (predicted == labels).sum().item()
    
print("正确率(%)：", 100 * correct / total)  # 打印正确率
print("样本总量",total)

每个类别的准确率

class_correct = list(0. for i in range(10))
class_total = list(0. for i in range(10))

with torch.no_grad():
    for data in testloader:
        images, labels = data
        outpus = net(images)
        _,predicted = torch.max(outpus,1)
        
        c = (predicted == labels).squeeze()  # 挤压维度：其中删除了维度为1的维度，返回true,false

        for i in range(4):
            label = labels[i]
            class_correct[label] += c[i].item()
            class_total[label] += 1
            
for i in range(10):
    print(classes[i], 100 * class_correct[i] / class_total[i])