Pytorch-实现ResNet-18并在Cifar-10数据集上进行验证

1.Pytorch上搭建ResNet-18

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


class ResBlk(nn.Module):
    """
    resnet block子模块
    """
    def __init__(self, ch_in, ch_out, stride=1):
   
        super(ResBlk, self).__init__()

        self.conv1 = nn.Conv2d(ch_in, ch_out, kernel_size=3, stride=stride, padding=1)
        self.bn1 = nn.BatchNorm2d(ch_out)
        self.conv2 = nn.Conv2d(ch_out, ch_out, kernel_size=3, stride=1, padding=1)
        self.bn2 = nn.BatchNorm2d(ch_out)

        self.extra = nn.Sequential()
        # 如果输入和输出的通道不一致，或其步长不为 1，需要将二者转成一致
        if ch_out != ch_in:
            # [b, ch_in, h, w] => [b, ch_out, h, w]
            self.extra = nn.Sequential(
                nn.Conv2d(ch_in, ch_out, kernel_size=1, stride=stride),
                nn.BatchNorm2d(ch_out)
            )

    def forward(self, x):
        
        out = F.relu(self.bn1(self.conv1(x)))
        out = self.bn2(self.conv2(out))
        
        out = self.extra(x) + out
        out = F.relu(out)        
        return out


class ResNet18(nn.Module):
    '''
    主模块
    '''
    def __init__(self):
        super(ResNet18, self).__init__()

        self.conv1 = nn.Sequential(
            nn.Conv2d(3, 64, kernel_size=3, stride=3, padding=0),
            nn.BatchNorm2d(64)
        )
        # followed 4 blocks       
        self.blk1 = ResBlk(64, 128, stride=2)          #[b, 64, h, w] => [b, 128, h ,w]
        self.blk2 = ResBlk(128, 256, stride=2)         #[b, 128, h, w] => [b, 256, h, w]
        self.blk3 = ResBlk(256, 512, stride=2)         #[b, 256, h, w] => [b, 512, h, w]
        self.blk4 = ResBlk(512, 512, stride=2)         #[b, 512, h, w] => [b, 512, h, w]

        self.outlayer = nn.Linear(512*1*1, 10)         #全连接层，总共10个分类

    def forward(self, x):
        x = F.relu(self.conv1(x))

        # [b, 64, h, w] => [b, 1024, h, w]
        x = self.blk1(x)
        x = self.blk2(x)
        x = self.blk3(x)
        x = self.blk4(x)

        x = F.adaptive_avg_pool2d(x, [1, 1])          #[b, 512, h, w] => [b, 512, 1, 1]
        x = x.view(x.size(0), -1)
        x = self.outlayer(x)

        return x

举个栗子测试一下：

if __name__ == '__main__':

    blk = ResBlk(64, 128, stride=4)
    tmp = torch.randn(2, 64, 32, 32)
    out = blk(tmp)
    print('block:', out.shape)                      #block: torch.Size([2, 128, 8, 8])

    x = torch.randn(2, 3, 32, 32)
    model = ResNet18()
    out = model(x)
    print('resnet:', out.shape)                     #resnet: torch.Size([2, 10])

2.训练Cifar-10数据集

所选数据集为Cifar-10，该数据集共有60000张带标签的彩色图像，这些图像尺寸32*32，分为10个类，每类6000张图。这里面有50000张用于训练，每个类5000张，另外10000用于测试，每个类1000张。

import  torch
from    torch.utils.data import DataLoader
from    torchvision import datasets,transforms 
from    torch import nn, optim

from    resnet import ResNet18


def main():
    batchsz = 128
    
    #训练集
    cifar_train = datasets.CIFAR10('cifar', True, transform=transforms.Compose([
        transforms.Resize((32, 32)),
        transforms.ToTensor(),
        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
    ]))
    cifar_train = DataLoader(cifar_train, batch_size=batchsz, shuffle=True)
    
    
    #测试集
    cifar_test = datasets.CIFAR10('cifar', False, transform=transforms.Compose([
        transforms.Resize((32, 32)),
        transforms.ToTensor(),
        transforms.Normalize(mean=[0.485, 0.456, 0.406], std=[0.229, 0.224, 0.225])
    ]))
    cifar_test = DataLoader(cifar_test, batch_size=batchsz, shuffle=True)


    x, label = iter(cifar_train).next()
    print('x:', x.shape, 'label:', label.shape)         #x: torch.Size([128, 3, 32, 32])  label: torch.Size([128])
    
    #定义模型-ResNet
    model = ResNet18()
    
    #定义损失函数和优化方式
    criteon = nn.CrossEntropyLoss()
    optimizer = optim.Adam(model.parameters(), lr=1e-3)
    print(model)
    
    #训练网络
    for epoch in range(1000):

        model.train()                                   #训练模式
        for batchidx, (x, label) in enumerate(cifar_train):
            #x: [b, 3, 32, 32]
            #label: [b]
            
            logits = model(x)                           #logits: [b, 10]
            loss = criteon(logits, label)               #标量

            optimizer.zero_grad()
            loss.backward()
            optimizer.step()

        print(epoch, 'loss:', loss.item())


        model.eval()                                    #测试模式
        with torch.no_grad():
            
            total_correct = 0                           #预测正确的个数
            total_num = 0
            for x, label in cifar_test:
                #x: [b, 3, 32, 32]
                #label: [b]
                              
                logits = model(x)                       #[b, 10]            
                pred = logits.argmax(dim=1)             #[b]
                
                # [b] vs [b] => scalar tensor
                correct = torch.eq(pred, label).float().sum().item()
                total_correct += correct
                total_num += x.size(0)
                
            acc = total_correct / total_num
            print(epoch, 'test acc:', acc)


if __name__ == '__main__':
    main()

迭代1000次，训练太久了，暂且输出前5次。

0 loss: 1.0912220478057861
0 test acc: 0.5583
1 loss: 0.8604468107223511
1 test acc: 0.6592
2 loss: 0.6625195145606995
2 test acc: 0.6827
3 loss: 0.7064175009727478
3 test acc: 0.6904
4 loss: 0.5687283277511597
4 test acc: 0.7059