第P10周：Pytorch实现车牌识别

• 🍨 本文为🔗365天深度学习训练营 中的学习记录博客
• 🍖 原作者：K同学啊

在之前的案例中，我们多是使用datasets.ImageFolder函数直接导入已经分类好的数据集形成Dataset，然后使用DataLoader加载Dataset，但是如果对无法分类的数据集，我们如何导入，并进行识别呢？

本周我将自定义一个MyDataset加载车牌数据集并完成车牌识别

🍺 基础要求：

1. 学习并理解本文

🍺 拔高要求：

1. 对单张车牌进行识别

🏡我的环境：

• 语言环境：Python3.8
• 编译器：Jupyter Lab
• 深度学习环境：
  • torch==2.2.2
  • torchvision==0.17.2

前期准备

• 如果设备上支持GPU就使用GPU,否则使用CPU
• Mac上的GPU使用mps

from torchvision.transforms import transforms
from torch.utils.data       import DataLoader
from torchvision            import datasets
import torchvision.models   as models
import torch.nn.functional  as F
import torch.nn             as nn
import torch,torchvision

import os,PIL,pathlib,warnings

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

# this ensures that the current MacOS version is at least 12.3+
print(torch.backends.mps.is_available())

# this ensures that the current current PyTorch installation was built with MPS activated.
print(torch.backends.mps.is_built())

# 设置硬件设备，如果有GPU则使用，没有则使用cpu
device = torch.device("mps" if torch.backends.mps.is_available() else "cpu")
device # # 使用的是GPU

True
True
device(type='mps')

一、导入数据

1.1. 获取类别名

import os,PIL,random,pathlib

data_dir = './data/p10/015_licence_plate/'
data_dir = pathlib.Path(data_dir)

data_paths  = list(data_dir.glob('*'))
classeNames = [str(path).split("/")[-1].split("_")[1].split(".")[0] for path in data_paths]
classeNames[:3]

['沪G1CE81', '云G86LR6', '鄂U71R9F']

data_paths     = list(data_dir.glob('*'))
data_paths_str = [str(path) for path in data_paths]
data_paths_str[:3]

['data/p10/015_licence_plate/000008250_沪G1CE81.jpg',
 'data/p10/015_licence_plate/000015082_云G86LR6.jpg',
 'data/p10/015_licence_plate/000004721_鄂U71R9F.jpg']

1.2. 数据可视化

import os,PIL,random,pathlib
import matplotlib.pyplot as plt

plt.figure(figsize=(14,5))
plt.suptitle("数据示例",fontsize=15)

for i in range(18):
    plt.subplot(3,6,i+1)
    
    # 显示图片
    images = plt.imread(data_paths_str[i])
    plt.imshow(images)

plt.show()

1.3. 标签数字化

import numpy as np

char_enum = ["京","沪","津","渝","冀","晋","蒙","辽","吉","黑","苏","浙","皖","闽","赣","鲁",\
              "豫","鄂","湘","粤","桂","琼","川","贵","云","藏","陕","甘","青","宁","新","军","使"]

number   = [str(i) for i in range(0, 10)]    # 0 到 9 的数字
alphabet = [chr(i) for i in range(65, 91)]   # A 到 Z 的字母

char_set       = char_enum + number + alphabet
char_set_len   = len(char_set)
label_name_len = len(classeNames[0])

# 将字符串数字化
def text2vec(text):
    vector = np.zeros([label_name_len, char_set_len])
    for i, c in enumerate(text):
        idx = char_set.index(c)
        vector[i][idx] = 1.0
    return vector

all_labels = [text2vec(i) for i in classeNames]

1.4. 加载数据文件

import os
import pandas as pd
from torchvision.io import read_image
from torch.utils.data import Dataset
import torch.utils.data as data
from PIL import Image

class MyDataset(data.Dataset):
    def __init__(self, all_labels, data_paths_str, transform):
        self.img_labels = all_labels      # 获取标签信息
        self.img_dir    = data_paths_str  # 图像目录路径
        self.transform  = transform       # 目标转换函数

    def __len__(self):
        return len(self.img_labels)

    def __getitem__(self, index):
        image    = Image.open(self.img_dir[index]).convert('RGB')#plt.imread(self.img_dir[index])  # 使用 torchvision.io.read_image 读取图像
        label    = self.img_labels[index]  # 获取图像对应的标签
        
        if self.transform:
            image = self.transform(image)
            
        return image, label  # 返回图像和标签

# 关于transforms.Compose的更多介绍可以参考：https://blog.csdn.net/qq_38251616/article/details/124878863
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 = MyDataset(all_labels, data_paths_str, train_transforms)
total_data

<__main__.MyDataset at 0x16cbaa430>

1.5. 划分数据

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

(10940, 2735)

train_loader = torch.utils.data.DataLoader(train_dataset,
                                           batch_size=16,
                                           shuffle=True)
test_loader = torch.utils.data.DataLoader(test_dataset,
                                          batch_size=16,
                                          shuffle=True)

print("The number of images in a training set is: ", len(train_loader)*16)
print("The number of images in a test set is: ", len(test_loader)*16)
print("The number of batches per epoch is: ", len(train_loader))

The number of images in a training set is:  10944
The number of images in a test set is:  2736
The number of batches per epoch is:  684

for X, y in test_loader:
    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([16, 3, 224, 224])
Shape of y:  torch.Size([16, 7, 69]) torch.float64

二、自建模型

2.1. 搭建模型

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
        """
        self.conv1 = nn.Conv2d(in_channels=3, out_channels=12, kernel_size=5, stride=1, padding=0)
        self.bn1 = nn.BatchNorm2d(12)
        self.conv2 = nn.Conv2d(in_channels=12, out_channels=12, kernel_size=5, stride=1, padding=0)
        self.bn2 = nn.BatchNorm2d(12)
        self.pool = nn.MaxPool2d(2,2)
        self.conv4 = nn.Conv2d(in_channels=12, out_channels=24, kernel_size=5, stride=1, padding=0)
        self.bn4 = nn.BatchNorm2d(24)
        self.conv5 = nn.Conv2d(in_channels=24, out_channels=24, kernel_size=5, stride=1, padding=0)
        self.bn5 = nn.BatchNorm2d(24)
        self.fc1 = nn.Linear(24*50*50, label_name_len*char_set_len)
        self.reshape = Reshape([label_name_len,char_set_len])

    def forward(self, x):
        x = F.relu(self.bn1(self.conv1(x)))      
        x = F.relu(self.bn2(self.conv2(x)))     
        x = self.pool(x)                        
        x = F.relu(self.bn4(self.conv4(x)))     
        x = F.relu(self.bn5(self.conv5(x)))  
        x = self.pool(x)                        
        x = x.view(-1, 24*50*50)
        x = self.fc1(x)
        
        # 最终reshape
        x = self.reshape(x)

        return x
    
# 定义Reshape层
class Reshape(nn.Module):
    def __init__(self, shape):
        super(Reshape, self).__init__()
        self.shape = shape

    def forward(self, x):
        return x.view(x.size(0), *self.shape)

print("Using {} device".format(device))

model = Network_bn().to(device)
model

Using mps 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=483, bias=True)
  (reshape): Reshape()
)

2.2. 查看模型详情

!pip install torchsummary

Looking in indexes: https://mirrors.aliyun.com/pypi/simple
Requirement already satisfied: torchsummary in /Users/henry/src/miniconda3/lib/python3.8/site-packages (1.5.1)

# 统计模型参数量以及其他指标
import torchsummary as summary

summary.summary(model.to("cpu"), (3, 224, 224))

----------------------------------------------------------------
        Layer (type)               Output Shape         Param #
================================================================
            Conv2d-1         [-1, 12, 220, 220]             912
       BatchNorm2d-2         [-1, 12, 220, 220]              24
            Conv2d-3         [-1, 12, 216, 216]           3,612
       BatchNorm2d-4         [-1, 12, 216, 216]              24
         MaxPool2d-5         [-1, 12, 108, 108]               0
            Conv2d-6         [-1, 24, 104, 104]           7,224
       BatchNorm2d-7         [-1, 24, 104, 104]              48
            Conv2d-8         [-1, 24, 100, 100]          14,424
       BatchNorm2d-9         [-1, 24, 100, 100]              48
        MaxPool2d-10           [-1, 24, 50, 50]               0
           Linear-11                  [-1, 483]      28,980,483
          Reshape-12                [-1, 7, 69]               0
================================================================
Total params: 29,006,799
Trainable params: 29,006,799
Non-trainable params: 0
----------------------------------------------------------------
Input size (MB): 0.57
Forward/backward pass size (MB): 26.56
Params size (MB): 110.65
Estimated Total Size (MB): 137.79
----------------------------------------------------------------

注意对比观察模型的输出[-1, 7, 69]，我们之前的网络结构输出都是[-1, 7]、[-1, 2]、[-1, 4]这样的二维数据，如果要求模型输出结果是多维数据，那么本案例将是很好的示例。

📮提问：[-1, 7, 69]中的-1是什么意思？

在神经网络中，如果我们不确定一个维度的大小，但是希望在计算中自动推断它，可以使用 -1。这个-1告诉 PyTorch 在计算中自动推断这个维度的大小，以确保其他维度的尺寸不变，并且能够保持张量的总大小不变。

例如，[-1, 7, 69]表示这个张量的形状是一个三维张量，其中第一个维度的大小是不确定的，第二维大小为7，第三大小分别为69。-1的作用是使得总的张量大小等于7 * 69，以适应实际的输入数据大小。

在实际的使用中，通常-1用在批处理维度上，因为在训练过程中，批处理大小可能会有所不同。使用-1可以使模型适应不同大小的批处理输入数据。

三、 训练模型

3.1. 优化器与损失函数

optimizer  = torch.optim.Adam(model.parameters(), 
                              lr=1e-4, 
                              weight_decay=0.0001)

loss_model = nn.CrossEntropyLoss()

本周任务之一：在下面的代码中我对loss进行了统计更新，请补充acc统计更新部分，即获取每一次测试的ACC值。

任务提示：pred.shape与y.shape是：[batch, 7, 69]，在进行acc计算时需注意～

from torch.autograd import Variable

def test(model, test_loader, loss_model):
    size = len(test_loader.dataset)
    num_batches = len(test_loader)
    
    model.eval()
    test_loss, correct = 0, 0
    with torch.no_grad():
        for X, y in test_loader:
            X, y = X.to(device), y.to(device)
            pred = model(X)

            test_loss += loss_model(pred, y).item()
            
    test_loss /= num_batches

    print(f"Avg loss: {test_loss:>8f} \n")
    return correct,test_loss

def train(model,train_loader,loss_model,optimizer):
    model=model.to(device)
    model.train()
    
    for i, (images, labels) in enumerate(train_loader, 0): #0是标起始位置的值。

        images = Variable(images.to(device))
        labels = Variable(labels.to(device))

        optimizer.zero_grad()
        outputs = model(images)

        loss = loss_model(outputs, labels)
        loss.backward()
        optimizer.step()

        if i % 1000 == 0:    
            print('[%5d] loss: %.3f' % (i, loss))

3.2. 模型的训练

test_acc_list  = []
test_loss_list = []
epochs = 30

for t in range(epochs):
    print(f"Epoch {t+1}\n--device-----------------------------")
    train(model,train_loader,loss_model,optimizer)
    test_acc,test_loss = test(model, test_loader, loss_model)
    test_acc_list.append(test_acc)
    test_loss_list.append(test_loss)
print("Done!")

Epoch 1
-------------------------------
[    0] loss: 0.213
Avg loss: 0.071063 

Epoch 2
-------------------------------
[    0] loss: 0.033
Avg loss: 0.057604 

......

Epoch 30
-------------------------------
[    0] loss: 0.014
Avg loss: 0.026364 

Done!

四、 结果分析

import numpy as np
import matplotlib.pyplot as plt

x = [i for i in range(1,31)]

plt.plot(x, test_loss_list, label="Loss", alpha=0.8)

plt.xlabel("Epoch")
plt.ylabel("Loss")

plt.legend()    
plt.show()