第P5周：运动鞋品牌识别

最新推荐文章于 2024-10-09 00:00:00 发布

计算机真好丸

最新推荐文章于 2024-10-09 00:00:00 发布

阅读量380

点赞数 3

文章标签：深度学习

本文链接：https://blog.csdn.net/weixin_52447211/article/details/141264973

版权

>- **🍨 本文为[🔗365天深度学习训练营](https://mp.weixin.qq.com/s/Z9yL_wt7L8aPOr9Lqb1K3w) 中的学习记录博客**
>- **🍖 原作者：[K同学啊](https://mtyjkh.blog.csdn.net/)**

一、前期准备

1.设置GPU

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='cpu')

2.导入数据

import os, PIL, random, pathlib

data_dir = './5-data/'
data_dir = pathlib.Path(data_dir)

data_paths = list(data_dir.glob('*'))
classeNames = [str(path).split("/")[1] for path in data_paths]
classeNames

['.DS_Store', 'test', 'train']

train_transforms = transforms.Compose([
    transforms.Resize([224,224]),          ## 将输入图片resize成统一尺寸
    ## transforms.RandomHorizontalFlip(),   ## 随机水平翻转
    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-->转换为标准正态分布（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1
    
])

test_transform = transforms.Compose([
    transforms.Resize([224, 224]),           ## 将图片resize成统一尺寸
    transforms.ToTensor(),
    transforms.Normalize(
        mean=[0.485, 0.456, 0.406],
        std=[0.229, 0.224, 0.225])      
])

train_dataset = datasets.ImageFolder("./5-data/train/",transform=train_transforms)
test_dataset = datasets.ImageFolder("./5-data/test/", transform=train_transforms)

train_dataset.class_to_idx

{'adidas': 0, 'nike': 1}

batch_size = 32 

train_dl = torch.utils.data.DataLoader(train_dataset,
                                           batch_size=batch_size,
                                           shuffle=True,
                                           num_workers=1)
test_dl = torch.utils.data.DataLoader(test_dataset,
                                          batch_size=batch_size,
                                          shuffle=True,
                                          num_workers=1)

for X, y in test_dl:
    print("Shape of X [N, C, H, W]: ", X.shape)
    print("Shape of y: ", y.shape, y.dtype)
    break

Shape of X [N, C, H, W]:  torch.Size([32, 3, 224, 224])
Shape of y:  torch.Size([32]) torch.int64

二、构建简单的CNN网络

import torch.nn.functional as F
 
class Model(nn.Module):
    def __init__(self):
        super(Model, self).__init__()
        self.conv1 = nn.Sequential(
            nn.Conv2d(3, 12, kernel_size = 5, padding=0),       ## 12*220*220
            nn.BatchNorm2d(12),
            nn.ReLU())
        
        self.conv2 = nn.Sequential(
            nn.Conv2d(12, 12,kernel_size = 5, padding = 0),     ## 12*216*216
            nn.BatchNorm2d(12),
            nn.ReLU())
        
        self.pool3 = nn.Sequential(                           
            nn.MaxPool2d(2))                                     ## 12*108*108
        
        self.conv4 = nn.Sequential(
            nn.Conv2d(12, 24,kernel_size = 5, padding = 0),      ## 24*104*104
            nn.BatchNorm2d(24),
            nn.ReLU())
        
        self.conv5 = nn.Sequential(
            nn.Conv2d(24, 24,kernel_size = 5, padding = 0),      ## 24*100*100
            nn.BatchNorm2d(24),
            nn.ReLU())
        
        self.pool6 = nn.Sequential(
            nn.MaxPool2d(2))                                     ## 24*50*50
        
        self.dropout = nn.Sequential(
            nn.Dropout(0.2))
        
        self.fc = nn.Sequential(
            nn.Linear(24*50*50, len(classeNames)))
        
    def forward(self, x):
            
        batch_size = x.size(0)
        x = self.conv1(x)         ## 卷积-BN-激活
        x = self.conv2(x)         ## 卷积-BN-激活
        x = self.pool3(x)         ## 池化
        x = self.conv4(x)         ## 卷积-BN-激活
        x = self.conv5(x)         ## 卷积-BN-激活
        x = self.pool6(x)         ## 池化
        x = self.dropout(x)
        x = x.view(batch_size, -1)   ## flatten变成全连接网络需要的输入(batch, 24*50*50)===>(batch, -1), -1此处自动计算出的是24*50*50
        x = self.fc(x)
            
        return x
        
device = "cuda" if torch.cuda.is_available() else "cpu"
print("Using {} device".format(device))

model = Model().to(device)
model

Using cpu device





Model(
  (conv1): Sequential(
    (0): Conv2d(3, 12, kernel_size=(5, 5), stride=(1, 1))
    (1): BatchNorm2d(12, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
  )
  (conv2): Sequential(
    (0): Conv2d(12, 12, kernel_size=(5, 5), stride=(1, 1))
    (1): BatchNorm2d(12, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
  )
  (pool3): Sequential(
    (0): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (conv4): Sequential(
    (0): Conv2d(12, 24, kernel_size=(5, 5), stride=(1, 1))
    (1): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
  )
  (conv5): Sequential(
    (0): Conv2d(24, 24, kernel_size=(5, 5), stride=(1, 1))
    (1): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)
    (2): ReLU()
  )
  (pool6): Sequential(
    (0): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)
  )
  (dropout): Sequential(
    (0): Dropout(p=0.2, inplace=False)
  )
  (fc): Sequential(
    (0): Linear(in_features=60000, out_features=3, bias=True)
  )
)

三、训练模型

1.编写训练函数

## 训练循环

def train(dataloader, model, loss_fn, optimizer):
    size = len(dataloader.dataset)       ## 训练集的大小
    num_batches = len(dataloader)        ## 批次数目，(size/batch_size,向上取整)
    
    train_loss, train_acc = 0, 0         ## 初始化train_loss train_acc
    
    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)          ## 计算网络输和真实值之间的差距
        
        ## 反向传播
        optimizer.zero_grad()            ## grad属性归零              
        loss.backward()                  ## 反向传播
        optimizer.step()                 ## 每一步自动更新
        
        ## 记录acc与loss
        train_acc += (pred.argmax(1) == y).type(torch.float).sum().item()
        train_loss += loss.item()
        
    train_acc /= size
    train_loss /= num_batches
    
    return train_acc, train_loss

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

def test(dataloader, model, loss_fn):
    size = len(dataloader.dataset)       ## 测试集大小
    num_batches = len(dataloader)        ## 批次数目,(size/batch_size,向上取整)
    test_acc, test_loss = 0, 0           ## 初始化test_acc, test_loss
    
    ## 当不进行训练时，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loss
            target_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 /= size
    test_loss /= num_batches
    
    return test_acc, test_loss

4.设置动态学习率

def adjust_learning_rate(optimizer, epoch, start_lr):
    ## 每2个epoch衰减到原来的0.92
    lr = start_lr * (0.92 ** (epoch // 2))
    for param_group in optimizer.param_groups:
        param_group['lr'] = lr
        
learn_rate = 1e-4                ## 初始化学习率
optimizer = torch.optim.SGD(model.parameters(), lr = learn_rate)

5.正式训练

loss_fn = nn.CrossEntropyLoss()      ## 创建损失函数
epochs = 40 

train_loss = []
train_acc = []
test_loss = []
test_acc = []

for epoch in range(epochs):
    ## 更新学习率(使用自定义学习率时使用)
    adjust_learning_rate(optimizer, epoch, learn_rate)
    
    model.train()
    epoch_train_acc, epoch_train_loss = train(train_dl, model, loss_fn, optimizer)

    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)
    
    ## 获取当前学习率
    lr = optimizer.state_dict()['param_groups'][0]['lr']
    
    template = ('Epoch:{:2d}, Train_acc:{:.1f}%, Train_loss:{:.3f}, Test_acc:{:.1f}%, Test_loss:{:.3f}, Lr:{:.2E}')
    print(template.format(epoch+1, epoch_train_acc*100, epoch_train_loss, epoch_test_acc*100, epoch_test_loss, lr))
    
print('Done')

Epoch: 1, Train_acc:55.4%, Train_loss:0.826, Test_acc:51.3%, Test_loss:0.894, Lr:1.00E-04
Epoch: 2, Train_acc:61.8%, Train_loss:0.693, Test_acc:53.9%, Test_loss:0.726, Lr:1.00E-04
Epoch: 3, Train_acc:66.1%, Train_loss:0.640, Test_acc:60.5%, Test_loss:0.647, Lr:9.20E-05
Epoch: 4, Train_acc:69.9%, Train_loss:0.591, Test_acc:68.4%, Test_loss:0.617, Lr:9.20E-05
Epoch: 5, Train_acc:75.9%, Train_loss:0.548, Test_acc:59.2%, Test_loss:0.629, Lr:8.46E-05
Epoch: 6, Train_acc:75.9%, Train_loss:0.534, Test_acc:67.1%, Test_loss:0.591, Lr:8.46E-05
Epoch: 7, Train_acc:76.5%, Train_loss:0.516, Test_acc:68.4%, Test_loss:0.592, Lr:7.79E-05
Epoch: 8, Train_acc:79.9%, Train_loss:0.473, Test_acc:71.1%, Test_loss:0.597, Lr:7.79E-05
Epoch: 9, Train_acc:81.5%, Train_loss:0.470, Test_acc:72.4%, Test_loss:0.568, Lr:7.16E-05
Epoch:10, Train_acc:83.9%, Train_loss:0.439, Test_acc:71.1%, Test_loss:0.540, Lr:7.16E-05
Epoch:11, Train_acc:85.1%, Train_loss:0.433, Test_acc:71.1%, Test_loss:0.552, Lr:6.59E-05
Epoch:12, Train_acc:85.1%, Train_loss:0.415, Test_acc:75.0%, Test_loss:0.575, Lr:6.59E-05
Epoch:13, Train_acc:85.1%, Train_loss:0.409, Test_acc:71.1%, Test_loss:0.578, Lr:6.06E-05
Epoch:14, Train_acc:86.7%, Train_loss:0.393, Test_acc:72.4%, Test_loss:0.539, Lr:6.06E-05
Epoch:15, Train_acc:87.1%, Train_loss:0.388, Test_acc:71.1%, Test_loss:0.522, Lr:5.58E-05
Epoch:16, Train_acc:88.0%, Train_loss:0.379, Test_acc:72.4%, Test_loss:0.518, Lr:5.58E-05
Epoch:17, Train_acc:89.0%, Train_loss:0.367, Test_acc:76.3%, Test_loss:0.525, Lr:5.13E-05
Epoch:18, Train_acc:89.4%, Train_loss:0.367, Test_acc:75.0%, Test_loss:0.508, Lr:5.13E-05
Epoch:19, Train_acc:90.0%, Train_loss:0.351, Test_acc:71.1%, Test_loss:0.479, Lr:4.72E-05
Epoch:20, Train_acc:89.6%, Train_loss:0.341, Test_acc:75.0%, Test_loss:0.474, Lr:4.72E-05
Epoch:21, Train_acc:90.2%, Train_loss:0.347, Test_acc:71.1%, Test_loss:0.491, Lr:4.34E-05
Epoch:22, Train_acc:89.2%, Train_loss:0.340, Test_acc:72.4%, Test_loss:0.499, Lr:4.34E-05
Epoch:23, Train_acc:90.8%, Train_loss:0.332, Test_acc:76.3%, Test_loss:0.448, Lr:4.00E-05
Epoch:24, Train_acc:92.0%, Train_loss:0.326, Test_acc:75.0%, Test_loss:0.512, Lr:4.00E-05
Epoch:25, Train_acc:92.2%, Train_loss:0.324, Test_acc:75.0%, Test_loss:0.498, Lr:3.68E-05
Epoch:26, Train_acc:91.2%, Train_loss:0.325, Test_acc:77.6%, Test_loss:0.499, Lr:3.68E-05
Epoch:27, Train_acc:92.6%, Train_loss:0.308, Test_acc:78.9%, Test_loss:0.488, Lr:3.38E-05
Epoch:28, Train_acc:93.2%, Train_loss:0.307, Test_acc:77.6%, Test_loss:0.465, Lr:3.38E-05
Epoch:29, Train_acc:93.2%, Train_loss:0.307, Test_acc:76.3%, Test_loss:0.474, Lr:3.11E-05
Epoch:30, Train_acc:93.8%, Train_loss:0.307, Test_acc:76.3%, Test_loss:0.521, Lr:3.11E-05
Epoch:31, Train_acc:92.6%, Train_loss:0.305, Test_acc:76.3%, Test_loss:0.482, Lr:2.86E-05
Epoch:32, Train_acc:93.6%, Train_loss:0.301, Test_acc:76.3%, Test_loss:0.501, Lr:2.86E-05
Epoch:33, Train_acc:92.2%, Train_loss:0.313, Test_acc:77.6%, Test_loss:0.497, Lr:2.63E-05
Epoch:34, Train_acc:94.2%, Train_loss:0.297, Test_acc:77.6%, Test_loss:0.485, Lr:2.63E-05
Epoch:35, Train_acc:92.4%, Train_loss:0.295, Test_acc:77.6%, Test_loss:0.509, Lr:2.42E-05
Epoch:36, Train_acc:95.8%, Train_loss:0.289, Test_acc:78.9%, Test_loss:0.445, Lr:2.42E-05
Epoch:37, Train_acc:95.4%, Train_loss:0.281, Test_acc:77.6%, Test_loss:0.467, Lr:2.23E-05
Epoch:38, Train_acc:94.4%, Train_loss:0.288, Test_acc:77.6%, Test_loss:0.480, Lr:2.23E-05
Epoch:39, Train_acc:94.0%, Train_loss:0.279, Test_acc:77.6%, Test_loss:0.497, Lr:2.05E-05
Epoch:40, Train_acc:95.4%, Train_loss:0.274, Test_acc:80.3%, Test_loss:0.474, Lr:2.05E-05
Done

四、结果可视化

1.Loss与Accuracy图

import matplotlib.pyplot as plt
## 隐藏警告
import warnings
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 = 'Training Loss')
plt.plot(epochs_range, test_loss, label = 'Test Loss')
plt.legend(loc = 'upper right')
plt.title('Training and Validation Loss')
plt.show()

在这里插入图片描述

2.指定图片进行预测

from PIL import Image

classes = list(train_dataset.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(image_path='./5-data/test/adidas/1.jpg',
                  model = model,
                  transform = train_transforms,
                  classes = classes)

预测结果是:adidas

五、保存并加载模型

## 模型保存
PATH = './model.path'        ## 保存的参数文件名
torch.save(model.state_dict(), PATH)

## 将参数加载到model中
model.load_state_dict(torch.load(PATH, map_location=device))

<All keys matched successfully>

六、动态学习率

1.torch.optim.lr_scheduler.StepLR
等间隔动态调整方法，每经过step_size个epoch，做一次学习率decay，以gammma值为缩小倍数。

optimizer = torch.optim.SGD(net.parameters(), lr = 0.001)
scheduler = torch.optim.lr_scheduler.StepLR(optimizer, step_size = 5, gamma = 0.1) 
## optimizer(Optimizer): 是之前定义好的需要优化的优化器的实例名名
## step_size(int): 是学习率衰减的周期，经过每个epoch，做一次学习率decay
## gammma(float): 学习率衰减的乘法因子，Default:0.1

2.lr_scheduler.LambdaLR
根据自己定义的函数更新学习率

lambda1 = lambda epoch: (0.92 ** (epoch // 2))        ## 第二组参数调整方法
optimizer = torch.optim.SGD(model.parameters(), lr = learn_rate)
scheduler = torch.optim.lr_scheduler.LambdaLR(optimizer, lr_lambda = lambda1)       ## 选定调整方法
## lr_lambda(function): 更新学习率的函数

3.lr_scheduler.MultiStepLR
在特定的epoch中调整学习率

optimizer = torch.optim.SGD(net.parameters(), lr = 0.001)
scheduler = torch.optim.lr_scheduler.MultiStepLR(optimizer,
                                                 milestone = [2, 6, 15],     ## 调整学习率的epoch数
                                                 gamma = 0.1)

调用官方接口示例：

model = [Parameter(torch.randn(2, 2, requires_grad = True))]
optimizer = SGD(model, 0.1)
scheduler = ExponentialLR(optimizer, gamma = 0.9)

for epoch in range(20)
    for input, target in dataset:
        optimizer.zero_grad()
        output = model(input)
        loss = loss_fn(output, target)
        loss.backward()
        optimizer.step()
    scheduler.step()

七、改进
改进需要测试集准确度达到86%
改进尝试一：

def adjust_learning_rate(optimizer, epoch, start_lr):
    ## 微调动态学习率，由每2个epoch衰减到原来的0.92变为每2个epoch衰减到原来的0.9
    lr = start_lr * (0.9 ** (epoch // 2))
    for param_group in optimizer.param_groups:
        param_group['lr'] = lr
        
learn_rate = 1e-4                ## 初始化学习率
optimizer = torch.optim.Adam(model.parameters(), lr = learn_rate) ## 优化器改用Adam

得出结果：
在这里插入图片描述
通过观察发现Test_acc有所改观超过了84%，但是还是没有到86%，初步发现中间测试集有来回震荡。分析原因，有可能是因为学习率过大或者是因为学习率衰减速度过快（比如每2个 epoch 就降低），模型可能会过早进入一个局部最优解，而不能充分利用数据进行学习。

总结：
本周主要学习了动态学习率的设置，动态学习率即在训练过程中，根据某种策略自动调整学习率，这种方法可以使模型加速收敛和陷入局部最优解。
动态学习率主要有3种设置方法：
1.可以等间隔设置衰减，即在经过规定的epoch后根据设定的参数进行衰减。
2.可以完全根据自己设计的函数来调整学习率
3.可以在规定的epoch中调整学习率