ResNet18 結構分析

一、概述

ResNet（殘差神經網路）於 2015 年提出，是 ImageNet 圖像分類任務上的冠軍。其核心思想是通過引入跨層連接（shortcut connection）解決了深層神經網路訓練過程中梯度消失和梯度爆炸的問題。ResNet18 是 ResNet 的一個較小版本，共有 18 層，包括 16 層卷積神經網路層和 2 層全連接層。

二、ResNet18 細節

1.卷積層

ResNet18 的卷積層共包含16層，在 3×3 的卷積核後，採用 ReLU 激活函數，步長為1，不使用池化層。其中，前7層卷積沒有跨層連接，8-16 層使用了跨層連接。具體來說，8-16 層中每隔一個殘差塊就有一條跨層連接，用於將上一層特徵圖與下一層殘差塊的輸出相加。

2.殘差塊

ResNet18 的每個殘差塊包含 2-3 個卷積層，每個卷積層都跟隨著 Batch Normalization 層和 ReLU 激活函數。具體來說，第一個卷積層的卷積核大小為 3 × 3，第二個卷積層的卷積核大小也為 3 × 3。如果跨層連接存在，還需要添加一層 1 × 1 的卷積層用於調整維度。

3.全連接層

ResNet18 的最後兩層全連接層都含有 512 個神經元，倒數第二層使用 ReLU 激活函數，而最後一層使用 Softmax 函數產生對類的預測輸出。

三、代碼示例

1.定義 ResNet18 結構

import torch.nn as nn

class BasicBlock(nn.Module):
    expansion = 1

    def __init__(self, in_planes, planes, stride=1):
        super(BasicBlock, self).__init__()
        self.conv1 = nn.Conv2d(in_planes, planes, kernel_size=3, stride=stride, padding=1, bias=False)
        self.bn1 = nn.BatchNorm2d(planes)
        self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=1, padding=1, bias=False)
        self.bn2 = nn.BatchNorm2d(planes)

        self.shortcut = nn.Sequential()
        if stride != 1 or in_planes != self.expansion*planes:
            self.shortcut = nn.Sequential(
                nn.Conv2d(in_planes, self.expansion*planes, kernel_size=1, stride=stride, bias=False),
                nn.BatchNorm2d(self.expansion*planes)
            )

    def forward(self, x):
        out = nn.functional.relu(self.bn1(self.conv1(x)))
        out = self.bn2(self.conv2(out))
        out += self.shortcut(x)
        out = nn.functional.relu(out)
        return out

class ResNet18(nn.Module):
    def __init__(self, num_classes=10):
        super(ResNet18, self).__init__()
        self.in_planes = 64

        self.conv1 = nn.Conv2d(3, 64, kernel_size=3, stride=1, padding=1, bias=False)
        self.bn1 = nn.BatchNorm2d(64)
        self.layer1 = self._make_layer(64, 2, stride=1)
        self.layer2 = self._make_layer(128, 2, stride=2)
        self.layer3 = self._make_layer(256, 2, stride=2)
        self.layer4 = self._make_layer(512, 2, stride=2)
        self.linear = nn.Linear(512*BasicBlock.expansion, num_classes)

    def _make_layer(self, planes, num_blocks, stride):
        strides = [stride] + [1]*(num_blocks-1)
        layers = []
        for stride in strides:
            layers.append(BasicBlock(self.in_planes, planes, stride))
            self.in_planes = planes * BasicBlock.expansion
        return nn.Sequential(*layers)

    def forward(self, x):
        out = nn.functional.relu(self.bn1(self.conv1(x)))
        out = self.layer1(out)
        out = self.layer2(out)
        out = self.layer3(out)
        out = self.layer4(out)
        out = nn.functional.avg_pool2d(out, 4)
        out = out.view(out.size(0), -1)
        out = self.linear(out)
        return out

2.實例化 ResNet18 模型

resnet18 = ResNet18()

3.模型訓練

criterion = nn.CrossEntropyLoss()
optimizer = torch.optim.Adam(resnet18.parameters())

for epoch in range(num_epochs):
    running_loss = 0.0
    for i, data in enumerate(trainloader, 0):
        inputs, labels = data
        optimizer.zero_grad()
        outputs = resnet18(inputs)
        loss = criterion(outputs, labels)
        loss.backward()
        optimizer.step()
        
        running_loss += loss.item()
        if i % 2000 == 1999:    
            print('[%d, %5d] loss: %.3f' %
                  (epoch + 1, i + 1, running_loss / 2000))
            running_loss = 0.0