第P6周：VGG-16算法实现人脸识别

最新推荐文章于 2024-03-07 21:04:54 发布

骑猪玩狗

最新推荐文章于 2024-03-07 21:04:54 发布

阅读量201

点赞数 3

文章标签：算法 python 深度学习

本文链接：https://blog.csdn.net/a536723241/article/details/135069814

版权

>- **🍨 本文为[🔗365天深度学习训练营](https://mp.weixin.qq.com/s/rbOOmire8OocQ90QM78DRA) 中的学习记录博客**
>- **🍖 原作者：[K同学啊 | 接辅导、项目定制](https://mtyjkh.blog.csdn.net/)**

🍺要求：

保存训练过程中的最佳模型权重
调用官方的VGG-16网络框架

🍻拔高（可选）：
3. 测试集准确率达到60%（难度有点大，但是这个过程可以学到不少）
4. 手动搭建VGG-16网络框架

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,warnings

warnings.filterwarnings("ignore")             #忽略警告信息

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

2. 导入数据

import os,PIL,random,pathlib

data_dir = './6-data/'
data_dir = pathlib.Path(data_dir)

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

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("./6-data/",transform=train_transforms)
total_data

将图片进行随机水平翻转处理，减少过拟合

total_data.class_to_idx

3. 划分数据集

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_dataset, test_dataset

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

二、调用官方的VGG-16模型

VGG-16结构说明：

● 13个卷积层（Convolutional Layer），分别用blockX_convX表示；
● 3个全连接层（Fully connected Layer），用classifier表示；
● 5个池化层（Pool layer）。

VGG-16包含了16个隐藏层（13个卷积层和3个全连接层），故称为VGG-16

from torchvision.models import vgg16

device = "cuda" if torch.cuda.is_available() else "cpu"
print("Using {} device".format(device))
    
# 加载预训练模型，并且对模型进行微调
model = vgg16(pretrained = True).to(device) # 加载预训练的vgg16模型

for param in model.parameters():
    param.requires_grad = False # 冻结模型的参数，这样子在训练的时候只训练最后一层的参数

# 修改classifier模块的第6层（即：(6): Linear(in_features=4096, out_features=2, bias=True)）
# 注意查看我们下方打印出来的模型
model.classifier = nn.Sequential(
    nn.Linear(512 * 7 * 7, 1024),
    nn.ReLU(inplace=True),
    nn.Dropout(0.5,inplace=False),
    nn.Linear(1024,512),
    nn.ReLU(inplace=True),
    nn.Dropout(0.5),
    nn.Linear(512,len(classeNames))) # 修改vgg16模型中最后一层全连接层，输出目标类别个数
model.to(device)  
model

参考了@数分大佬的方法，1、减少了全连接层参数 2、增加dropout层，减少过拟合

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之间的差距，targets为真实值，计算二者差值即为损失
        
        # 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

2. 编写测试函数

def test (dataloader, model, loss_fn):
    size        = len(dataloader.dataset)  # 测试集的大小
    num_batches = len(dataloader)          # 批次数目, (size/batch_size，向上取整)
    test_loss, test_acc = 0, 0
    
    # 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

3. 设置动态学习率

learn_rate = 1e-3 # 初始学习率
# 调用官方动态学习率接口时使用
lambda1 = lambda epoch: 0.92 ** (epoch // 2)
optimizer = torch.optim.Adam(model.parameters(), lr=learn_rate)
scheduler = torch.optim.lr_scheduler.LambdaLR(optimizer, lr_lambda=lambda1) #选定调整方法

将初始学习率提高，这一步对准确率提升最大；同时将学习率衰减速度改为每2回合衰减为原来的0.92

4. 正式训练

import copy

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

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

best_acc = 0    # 设置一个最佳准确率，作为最佳模型的判别指标

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)
    scheduler.step() # 更新学习率（调用官方动态学习率接口时使用）
    
    model.eval()
    epoch_test_acc, epoch_test_loss = test(test_dl, model, loss_fn)
    
    # 保存最佳模型到 best_model
    if epoch_test_acc > best_acc:
        best_acc   = epoch_test_acc
        best_model = copy.deepcopy(model)
    
    train_acc.append(epoch_train_acc)
    train_loss.append(epoch_train_loss)
    test_acc.append(epoch_test_acc)
    test_loss.append(epoch_test_loss)
    
    # 获取当前的学习率
    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))
    
# 保存最佳模型到文件中
PATH = 'F:/365data/best_model.pth'  # 保存的参数文件名
torch.save(model.state_dict(), PATH)

print('Done')

Epoch: 1, Train_acc:16.8%, Train_loss:2.659, Test_acc:34.4%, Test_loss:2.003, Lr:1.00E-03
Epoch: 2, Train_acc:37.6%, Train_loss:1.822, Test_acc:42.5%, Test_loss:1.604, Lr:9.20E-04
Epoch: 3, Train_acc:51.2%, Train_loss:1.435, Test_acc:49.7%, Test_loss:1.492, Lr:9.20E-04
Epoch: 4, Train_acc:63.1%, Train_loss:1.112, Test_acc:53.1%, Test_loss:1.347, Lr:8.46E-04
Epoch: 5, Train_acc:73.0%, Train_loss:0.791, Test_acc:50.3%, Test_loss:1.361, Lr:8.46E-04
Epoch: 6, Train_acc:79.5%, Train_loss:0.617, Test_acc:56.7%, Test_loss:1.392, Lr:7.79E-04
Epoch: 7, Train_acc:83.3%, Train_loss:0.477, Test_acc:53.1%, Test_loss:1.396, Lr:7.79E-04
Epoch: 8, Train_acc:88.3%, Train_loss:0.359, Test_acc:58.3%, Test_loss:1.336, Lr:7.16E-04
Epoch: 9, Train_acc:90.1%, Train_loss:0.320, Test_acc:55.6%, Test_loss:1.384, Lr:7.16E-04
Epoch:10, Train_acc:92.4%, Train_loss:0.231, Test_acc:60.3%, Test_loss:1.421, Lr:6.59E-04
Epoch:11, Train_acc:94.0%, Train_loss:0.193, Test_acc:57.2%, Test_loss:1.476, Lr:6.59E-04
Epoch:12, Train_acc:93.6%, Train_loss:0.183, Test_acc:53.3%, Test_loss:1.563, Lr:6.06E-04
Epoch:13, Train_acc:95.4%, Train_loss:0.159, Test_acc:57.5%, Test_loss:1.479, Lr:6.06E-04
Epoch:14, Train_acc:94.9%, Train_loss:0.149, Test_acc:54.4%, Test_loss:1.519, Lr:5.58E-04
Epoch:15, Train_acc:95.2%, Train_loss:0.146, Test_acc:53.9%, Test_loss:1.578, Lr:5.58E-04
Epoch:16, Train_acc:96.8%, Train_loss:0.117, Test_acc:56.7%, Test_loss:1.583, Lr:5.13E-04
Epoch:17, Train_acc:97.3%, Train_loss:0.088, Test_acc:55.6%, Test_loss:1.696, Lr:5.13E-04
Epoch:18, Train_acc:97.2%, Train_loss:0.100, Test_acc:59.4%, Test_loss:1.556, Lr:4.72E-04
Epoch:19, Train_acc:97.7%, Train_loss:0.079, Test_acc:55.6%, Test_loss:1.673, Lr:4.72E-04
Epoch:20, Train_acc:97.8%, Train_loss:0.067, Test_acc:55.8%, Test_loss:1.601, Lr:4.34E-04
Epoch:21, Train_acc:98.0%, Train_loss:0.058, Test_acc:59.4%, Test_loss:1.466, Lr:4.34E-04
Epoch:22, Train_acc:98.8%, Train_loss:0.042, Test_acc:59.2%, Test_loss:1.754, Lr:4.00E-04
Epoch:23, Train_acc:99.0%, Train_loss:0.036, Test_acc:58.1%, Test_loss:1.755, Lr:4.00E-04
Epoch:24, Train_acc:98.7%, Train_loss:0.037, Test_acc:57.2%, Test_loss:1.717, Lr:3.68E-04
Epoch:25, Train_acc:98.5%, Train_loss:0.043, Test_acc:56.7%, Test_loss:1.772, Lr:3.68E-04
Epoch:26, Train_acc:98.8%, Train_loss:0.038, Test_acc:58.3%, Test_loss:1.601, Lr:3.38E-04
Epoch:27, Train_acc:99.1%, Train_loss:0.024, Test_acc:58.6%, Test_loss:1.550, Lr:3.38E-04
Epoch:28, Train_acc:99.2%, Train_loss:0.032, Test_acc:56.9%, Test_loss:1.710, Lr:3.11E-04
Epoch:29, Train_acc:98.9%, Train_loss:0.036, Test_acc:57.5%, Test_loss:1.753, Lr:3.11E-04
Epoch:30, Train_acc:99.3%, Train_loss:0.029, Test_acc:56.1%, Test_loss:1.781, Lr:2.86E-04
Epoch:31, Train_acc:99.5%, Train_loss:0.019, Test_acc:61.4%, Test_loss:1.738, Lr:2.86E-04
Epoch:32, Train_acc:99.1%, Train_loss:0.034, Test_acc:58.1%, Test_loss:1.826, Lr:2.63E-04
Epoch:33, Train_acc:99.1%, Train_loss:0.044, Test_acc:57.5%, Test_loss:1.852, Lr:2.63E-04
Epoch:34, Train_acc:99.5%, Train_loss:0.020, Test_acc:57.8%, Test_loss:1.804, Lr:2.42E-04
Epoch:35, Train_acc:99.5%, Train_loss:0.020, Test_acc:57.2%, Test_loss:1.745, Lr:2.42E-04
Epoch:36, Train_acc:99.2%, Train_loss:0.030, Test_acc:56.1%, Test_loss:1.732, Lr:2.23E-04
Epoch:37, Train_acc:98.8%, Train_loss:0.034, Test_acc:58.9%, Test_loss:1.762, Lr:2.23E-04
Epoch:38, Train_acc:99.4%, Train_loss:0.021, Test_acc:57.2%, Test_loss:1.778, Lr:2.05E-04
Epoch:39, Train_acc:99.5%, Train_loss:0.016, Test_acc:58.1%, Test_loss:1.933, Lr:2.05E-04
Epoch:40, Train_acc:99.3%, Train_loss:0.021, Test_acc:58.9%, Test_loss:1.766, Lr:1.89E-04
Done

最终测试集准确率达到了61.4%

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(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}')

# 预测训练集中的某张照片
predict_one_image(image_path='./6-data/Angelina Jolie/001_fe3347c0.jpg', 
                  model=model, 
                  transform=train_transforms, 
                  classes=classes)

3. 模型评估

best_model.eval()
epoch_test_acc, epoch_test_loss = test(test_dl, best_model, loss_fn)

epoch_test_acc, epoch_test_loss

# 查看是否与我们记录的最高准确率一致
epoch_test_acc