Task03：字符识别模型

0. CNN原理和发展

CNN由卷积（convolution）、池化（pooling）、非线性激活函数（non-linear activation function）和全连接层（fully connected layer）构成。通过多次卷积和池化，最后一层将输入的图像像素映射为具体的输出。如在分类任务中会转换为不同类别的概率输出，然后计算真实标签与CNN模型的预测结果的差异，并通过反向传播更新每层的参数，并在更新完成后再次前向传播，如此反复直到训练完成。
与传统机器学习模型相比，CNN具有一种端到端（End to End）的思路。在CNN训练的过程中是直接从图像像素到最终的输出，并不涉及到具体的特征提取和构建模型的过程，也不需要人工的参与。

具体概念的细节可以在这里找到。

卷积过程，感受野逐渐增大

CNN的网络结构和参数趋于更深和更多方向发展，出现了很多经典的网络模型，例如LeNet、AlexNet、VGG、Inception、ResNet等等。

模型的具体结构和PyTorch构建的代码可以在这里找到。

1. PyTorch构建CNN字符分类器

一个示例分类器：包括两个卷积层用于提取特征，并联6个全连接层用于分类。

import torch
torch.manual_seed(0)
torch.backends.cudnn.deterministic = False
torch.backends.cudnn.benchmark = True

import torchvision.models as models
import torchvision.transforms as transforms
import torchvision.datasets as datasets
import torch.nn as nn
import torch.nn.functional as F
import torch.optim as optim
from torch.autograd import Variable
from torch.utils.data.dataset import Dataset

class SVHN_Model1(nn.Module):
    def __init__(self):
        super(SVHN_Model1, self).__init__()
        # CNN提取特征模块
        self.cnn = nn.Sequential(
            # 3通道输入, 16通道输出, 3x3 kernel, 右下步幅均为2
            nn.Conv2d(3, 16, kernel_size=(3, 3), stride=(2, 2)),
            nn.ReLU(),  
            nn.MaxPool2d(2),
            nn.Conv2d(16, 32, kernel_size=(3, 3), stride=(2, 2)),
            nn.ReLU(), 
            nn.MaxPool2d(2),
        )
        # 此处将用于分类的全连接层设置为6个，对每位数字进行分类
        self.fc1 = nn.Linear(32*3*7, 11)
        self.fc2 = nn.Linear(32*3*7, 11)
        self.fc3 = nn.Linear(32*3*7, 11)
        self.fc4 = nn.Linear(32*3*7, 11)
        self.fc5 = nn.Linear(32*3*7, 11)
        self.fc6 = nn.Linear(32*3*7, 11)
    
    def forward(self, img):        
        feat = self.cnn(img)
        feat = feat.view(feat.shape[0], -1)
        c1 = self.fc1(feat)
        c2 = self.fc2(feat)
        c3 = self.fc3(feat)
        c4 = self.fc4(feat)
        c5 = self.fc5(feat)
        c6 = self.fc6(feat)
        return c1, c2, c3, c4, c5, c6

model = SVHN_Model1()

训练代码如下

# 损失函数
criterion = nn.CrossEntropyLoss()
# 优化器
optimizer = torch.optim.Adam(model.parameters(), 0.005)

loss_plot, c0_plot = [], []
# 迭代10个Epoch
for epoch in range(10):
    for data in train_loader:
        c0, c1, c2, c3, c4, c5 = model(data[0])
        loss = criterion(c0, data[1][:, 0]) + \
                criterion(c1, data[1][:, 1]) + \
                criterion(c2, data[1][:, 2]) + \
                criterion(c3, data[1][:, 3]) + \
                criterion(c4, data[1][:, 4]) + \
                criterion(c5, data[1][:, 5])
        loss /= 6
        optimizer.zero_grad()
        loss.backward()
        optimizer.step()
        
        loss_plot.append(loss.item())
        c0_plot.append((c0.argmax(1) == data[1][:, 0]).sum().item()*1.0 / c0.shape[0])
        
    print(epoch)

- 迁移学习：使用ImageNet与训练模型构建SVHN数据集的CNN分类器

class SVHN_Model2(nn.Module):
    def __init__(self):
        super(SVHN_Model1, self).__init__()

        ################################################################
        model_conv = models.resnet18(pretrained=True)                 ##
        model_conv.avgpool = nn.AdaptiveAvgPool2d(1)                  ##                          
        model_conv = nn.Sequential(*list(model_conv.children())[:-1]) ##
        ################################################################
        self.cnn = model_conv
        
        self.fc1 = nn.Linear(512, 11)
        self.fc2 = nn.Linear(512, 11)
        self.fc3 = nn.Linear(512, 11)
        self.fc4 = nn.Linear(512, 11)
        self.fc5 = nn.Linear(512, 11)
    
    def forward(self, img):        
        feat = self.cnn(img)
        # print(feat.shape)
        feat = feat.view(feat.shape[0], -1)
        c1 = self.fc1(feat)
        c2 = self.fc2(feat)
        c3 = self.fc3(feat)
        c4 = self.fc4(feat)
        c5 = self.fc5(feat)
        return c1, c2, c3, c4, c5