Resnet网络实战之Fashion-MNIST数据集

残差网络的由来（Resnet）

 在实践中，添加过多的层后训练误差往往不降反升。即使利用批量归一化带来的数值稳定性使训练深层模型更加容易，该问题仍然存在。针对这一问题，何恺明等人提出了残差网络（ResNet）。它在2015年的ImageNet图像识别挑战赛夺魁，并深刻影响了后来的深度神经网络的设计。下面通过Fashion-MNIST数据集来了解一下ResNet-18网络。

普通的网络结构与加入残差连接的网络结构对比

 设输入为x。假设我们希望学出的理想映射为f(x)，从而作为图最上方激活函数的输入。左图虚线框中的部分需要直接拟合出该映射f(x)，而右图虚线框中的部分则需要拟合出有关恒等映射的残差映射f(x)−x。残差映射在实际中往往更容易优化。我们只需将图中右图虚线框内上方的加权运算（如仿射）的权重和偏差参数学成0，那么f(x)即为恒等映射。实际中，当理想映射f(x)极接近于恒等映射时，残差映射也易于捕捉恒等映射的细微波动。右图也是ResNet的基础块，即残差块（residual block）。在残差块中，输入可通过跨层的数据线路更快地向前传播。

残差块的代码实现

 ResNet沿用了VGG全3×3卷积层的设计。残差块里首先有2个有相同输出通道数的3×3卷积层。每个卷积层后接一个批量归一化层和ReLU激活函数。然后我们将输入跳过这两个卷积运算后直接加在最后的ReLU激活函数前。这样的设计要求两个卷积层的输出与输入形状一样，从而可以相加。如果想改变通道数，就需要引入一个额外的1×1卷积层来将输入变换成需要的形状后再做相加运算。
残差块的实现如下。它可以设定输出通道数、是否使用额外的1×1卷积层来修改通道数以及卷积层的步幅。

import time
import torch
from torch import nn, optim
import torch.nn.functional as F

import sys
device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')
class Residual(nn.Module):  
    def __init__(self, in_channels, out_channels, use_1x1conv=False, stride=1):
        super(Residual, self).__init__()
        self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1, stride=stride)
        self.conv2 = nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1)
        if use_1x1conv:
            self.conv3 = nn.Conv2d(in_channels, out_channels, kernel_size=1, stride=stride)
        else:
            self.conv3 = None
        self.bn1 = nn.BatchNorm2d(out_channels)
        self.bn2 = nn.BatchNorm2d(out_channels)

    def forward(self, X):
        Y = F.relu(self.bn1(self.conv1(X)))
        Y = self.bn2(self.conv2(Y))
        if self.conv3:
            X = self.conv3(X)
        return F.relu(Y + X)

Resnet模型

 ResNet的前两层跟GoogLeNet中的一样：在输出通道数为64、步幅为2的7×7卷积层后接步幅为2的3×3的最大池化层。不同之处在于ResNet每个卷积层后增加的批量归一化层

net = nn.Sequential(
        nn.Conv2d(1, 64, kernel_size=7, stride=2, padding=3),
        nn.BatchNorm2d(64), 
        nn.ReLU(),
        nn.MaxPool2d(kernel_size=3, stride=2, padding=1))

 ResNet则使用4个由残差块组成的模块，每个模块使用若干个同样输出通道数的残差块。第一个模块的通道数同输入通道数一致。由于之前已经使用了步幅为2的最大池化层，所以无须减小高和宽。之后的每个模块在第一个残差块里将上一个模块的通道数翻倍，并将高和宽减半。
下面我们来实现这个模块。注意，这里对第一个模块做了特别处理（即第一个模块的in_channels == out_channels）。

def resnet_block(in_channels, out_channels, num_residuals, first_block=False):
    if first_block:
        assert in_channels == out_channels # 第一个模块的通道数同输入通道数一致
    blk = []
    for i in range(num_residuals):
        if i == 0 and not first_block:
            blk.append(Residual(in_channels, out_channels, use_1x1conv=True, stride=2))
        else:
            blk.append(Residual(out_channels, out_channels))
    return nn.Sequential(*blk)

 接着我们为ResNet加入所有残差块。这里每个模块使用两个残差块，即[2, 2, 2, 2]。

net.add_module("resnet_block1", resnet_block(64, 64, 2, first_block=True))
net.add_module("resnet_block2", resnet_block(64, 128, 2))
net.add_module("resnet_block3", resnet_block(128, 256, 2))
net.add_module("resnet_block4", resnet_block(256, 512, 2))

 最后，与GoogLeNet一样，加入全局平均池化层后接上全连接层输出。

net.add_module("global_avg_pool", GlobalAvgPool2d()) # GlobalAvgPool2d的输出: (Batch, 512, 1, 1)
net.add_module("fc", nn.Sequential(FlattenLayer(), nn.Linear(512, 10))) 

 这里每个模块里有4个卷积层（不计算1×1卷积层），加上最开始的卷积层和最后的全连接层，共计18层。这个模型通常也被称为ResNet-18。通过配置不同的通道数和模块里的残差块数可以得到不同的ResNet模型，例如更深的含152层的ResNet-152。虽然ResNet的主体架构跟GoogLeNet的类似，但ResNet结构更简单，修改也更方便。这些因素都导致了ResNet迅速被广泛使用。
在训练ResNet之前，我们来观察一下输入形状在ResNet不同模块之间的变化。

X = torch.rand((1, 1, 224, 224))
for name, layer in net.named_children():
    X = layer(X)
    print(name, ' output shape:\t', X.shape)

结果：

获取数据和训练模型

 下面我们在Fashion-MNIST数据集上训练ResNet。

batch_size = 64#批次可以设置大点，256
# 如出现“out of memory”的报错信息，可减小batch_size或resize
train_iter, test_iter = load_data_fashion_mnist(batch_size, resize=96)
lr, num_epochs = 0.001, 5
optimizer = torch.optim.Adam(net.parameters(), lr=lr)
train_ch5(net, train_iter, test_iter, batch_size, optimizer, device, num_epochs)

 结果:

完整代码

import time
import torch
from torch import nn, optim
import torch.nn.functional as F
import torchvision
device = torch.device('cuda:1' if torch.cuda.is_available() else 'cpu')

class Residual(nn.Module):  
    def __init__(self, in_channels, out_channels, use_1x1conv=False, stride=1):
        super(Residual, self).__init__()
        self.conv1 = nn.Conv2d(in_channels, out_channels, kernel_size=3, padding=1, stride=stride)
        self.conv2 = nn.Conv2d(out_channels, out_channels, kernel_size=3, padding=1)
        if use_1x1conv:
            self.conv3 = nn.Conv2d(in_channels, out_channels, kernel_size=1, stride=stride)
        else:
            self.conv3 = None
        self.bn1 = nn.BatchNorm2d(out_channels)
        self.bn2 = nn.BatchNorm2d(out_channels)

    def forward(self, X):
        Y = F.relu(self.bn1(self.conv1(X)))
        Y = self.bn2(self.conv2(Y))
        if self.conv3:
            X = self.conv3(X)
        return F.relu(Y + X)

def resnet_block(in_channels, out_channels, num_residuals, first_block=False):
    if first_block:
        assert in_channels == out_channels # 第一个模块的通道数同输入通道数一致
    blk = []
    for i in range(num_residuals):
        if i == 0 and not first_block:
            blk.append(Residual(in_channels, out_channels, use_1x1conv=True, stride=2))
        else:
            blk.append(Residual(out_channels, out_channels))
    return nn.Sequential(*blk)


def load_data_fashion_mnist(batch_size, resize=None, root='~/Datasets/FashionMNIST'):
    """Download the fashion mnist dataset and then load into memory."""
    trans = []
    if resize:
        trans.append(torchvision.transforms.Resize(size=resize))
    trans.append(torchvision.transforms.ToTensor())

    transform = torchvision.transforms.Compose(trans)
    mnist_train = torchvision.datasets.FashionMNIST(root=root, train=True, download=False, transform=transform)
    mnist_test = torchvision.datasets.FashionMNIST(root=root, train=False, download=False, transform=transform)
    if sys.platform.startswith('win'):
        num_workers = 0  # 0表示不用额外的进程来加速读取数据
    else:
        num_workers = 4
    train_iter = torch.utils.data.DataLoader(mnist_train, batch_size=batch_size, shuffle=True, num_workers=num_workers)
    test_iter = torch.utils.data.DataLoader(mnist_test, batch_size=batch_size, shuffle=False, num_workers=num_workers)

    return train_iter, test_iter

class GlobalAvgPool2d(nn.Module):
    # 全局平均池化层可通过将池化窗口形状设置成输入的高和宽实现
    def __init__(self):
        super(GlobalAvgPool2d, self).__init__()
    def forward(self, x):
        return F.avg_pool2d(x, kernel_size=x.size()[2:])

class FlattenLayer(torch.nn.Module):
    def __init__(self):
        super(FlattenLayer, self).__init__()
    def forward(self, x): # x shape: (batch, *, *, ...)
        return x.view(x.shape[0], -1)


def evaluate_accuracy(data_iter, net, device=None):
    if device is None and isinstance(net, torch.nn.Module):
        # 如果没指定device就使用net的device
        device = list(net.parameters())[0].device
    acc_sum, n = 0.0, 0
    with torch.no_grad():
        for X, y in data_iter:
            if isinstance(net, torch.nn.Module):
                net.eval() # 评估模式, 这会关闭dropout
                acc_sum += (net(X.to(device)).argmax(dim=1) == y.to(device)).float().sum().cpu().item()
                net.train() # 改回训练模式
            else: # 自定义的模型, 3.13节之后不会用到, 不考虑GPU
                if('is_training' in net.__code__.co_varnames): # 如果有is_training这个参数
                    # 将is_training设置成False
                    acc_sum += (net(X, is_training=False).argmax(dim=1) == y).float().sum().item()
                else:
                    acc_sum += (net(X).argmax(dim=1) == y).float().sum().item()
            n += y.shape[0]
    return acc_sum / n

def train_ch5(net, train_iter, test_iter, batch_size, optimizer, device, num_epochs):
    net = net.to(device)
    print("training on ", device)
    loss = torch.nn.CrossEntropyLoss()
    for epoch in range(num_epochs):
        train_l_sum, train_acc_sum, n, batch_count, start = 0.0, 0.0, 0, 0, time.time()
        for X, y in train_iter:
            X = X.to(device)
            y = y.to(device)
            y_hat = net(X)
            l = loss(y_hat, y)
            optimizer.zero_grad()
            l.backward()
            optimizer.step()
            train_l_sum += l.cpu().item()
            train_acc_sum += (y_hat.argmax(dim=1) == y).sum().cpu().item()
            n += y.shape[0]
            batch_count += 1
        test_acc = evaluate_accuracy(test_iter, net)
        print('epoch %d, loss %.4f, train acc %.3f, test acc %.3f, time %.1f sec'
              % (epoch + 1, train_l_sum / batch_count, train_acc_sum / n, test_acc, time.time() - start))
net = nn.Sequential(
        nn.Conv2d(1, 64, kernel_size=7, stride=2, padding=3),
        nn.BatchNorm2d(64),
        nn.ReLU(),
        nn.MaxPool2d(kernel_size=3, stride=2, padding=1))
net.add_module("resnet_block1", resnet_block(64, 64, 2, first_block=True))

net.add_module("resnet_block2", resnet_block(64, 128, 2))
net.add_module("resnet_block3", resnet_block(128, 256, 2))
net.add_module("resnet_block4", resnet_block(256, 512, 2))
net.add_module("global_avg_pool", GlobalAvgPool2d()) # GlobalAvgPool2d的输出: (Batch, 512, 1, 1)

net.add_module("fc", nn.Sequential(FlattenLayer(), nn.Linear(512, 10)))
batch_size = 64
# 如出现“out of memory”的报错信息，可减小batch_size或resize
train_iter, test_iter = load_data_fashion_mnist(batch_size, resize=96)
lr, num_epochs = 0.001, 5
optimizer = torch.optim.Adam(net.parameters(), lr=lr)
train_ch5(net, train_iter, test_iter, batch_size, optimizer, device, num_epochs)