https://github.com/pytorch/vision/blob/9a481d0bec2700763a799ff148fe2e083b575441/torchvision/models/resnet.py
各種ResNet網(wǎng)絡(luò)是由BasicBlock或者bottleneck構(gòu)成的，它們是構(gòu)成深度殘差網(wǎng)絡(luò)的基本模塊

ResNet主體

ResNet的大部分各種結(jié)構(gòu)是1層conv+4個(gè)block+1層fc

class ResNet(nn.Module):

    def __init__(self, block, layers, zero_init_residual=False):
        super(ResNet, self).__init__()
        self.inplanes = 64
        self.conv1 = nn.Conv2d(3, 64, kernel_size=7, stride=2, padding=3,
                               bias=False)
        self.bn1 = nn.BatchNorm2d(64)
        self.relu = nn.ReLU(inplace=True)
        self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
        self.layer1 = self._make_layer(block, 64, layers[0])
        self.layer2 = self._make_layer(block, 128, layers[1], stride=2)
        self.layer3 = self._make_layer(block, 256, layers[2], stride=2)
        self.layer4 = self._make_layer(block, 512, layers[3], stride=2)
		self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
        self.fc = nn.Linear(512 * block.expansion, num_classes)
        for m in self.modules():
            if isinstance(m, nn.Conv2d):
                nn.init.kaiming_normal_(m.weight, mode='fan_out', nonlinearity='relu')
            elif isinstance(m, nn.BatchNorm2d):
                nn.init.constant_(m.weight, 1)
                nn.init.constant_(m.bias, 0)

        # Zero-initialize the last BN in each residual branch,
        # so that the residual branch starts with zeros, and each residual block behaves like an identity.
        # This improves the model by 0.2~0.3% according to https://arxiv.org/abs/1706.02677
        if zero_init_residual:
            for m in self.modules():
                if isinstance(m, Bottleneck):
                    nn.init.constant_(m.bn3.weight, 0)
                elif isinstance(m, BasicBlock):
                    nn.init.constant_(m.bn2.weight, 0)

    def _make_layer(self, block, planes, blocks, stride=1):
        downsample = None
        if stride != 1 or self.inplanes != planes * block.expansion:
         # normly happened when stride = 2
            downsample = nn.Sequential(
                conv1x1(self.inplanes, planes * block.expansion, stride),
                nn.BatchNorm2d(planes * block.expansion),
            )

        layers = []
        layers.append(block(self.inplanes, planes, stride, downsample))
        self.inplanes = planes * block.expansion
        for _ in range(1, blocks):  
        # only the first block need downsample thus there is no downsample and stride = 2
            layers.append(block(self.inplanes, planes))

        return nn.Sequential(*layers)

    def forward(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        x = self.maxpool(x)

        c2 = self.layer1(x)
        c3 = self.layer2(c2)
        c4 = self.layer3(c3)
        c5 = self.layer4(c4)
        x = self.avgpool(x)
        x = x.view(x.size(0), -1)
        x = self.fc(x)
        return c5

需要注意的是最后的avgpool是全局的平均池化

BasicBlock

class BasicBlock(nn.Module):
    expansion = 1

    def __init__(self, inplanes, planes, stride=1, downsample=None):
    	# here planes names channel number
        super(BasicBlock, self).__init__()
        self.conv1 = conv3x3(inplanes, planes, stride)
        self.bn1 = nn.BatchNorm2d(planes)
        self.relu = nn.ReLU(inplace=True)
        self.conv2 = conv3x3(planes, planes)
        self.bn2 = nn.BatchNorm2d(planes)
        self.downsample = downsample
        self.stride = stride

    def forward(self, x):
        identity = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn2(out)

        if self.downsample is not None:
            identity = self.downsample(x)

        out += identity
        out = self.relu(out)

        return out

ResNet18

對(duì)應(yīng)的就是[2,2,2,2]

def resnet18(pretrained=False, **kwargs):
    """Constructs a ResNet-18 model.

    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet
    """
    model = ResNet(BasicBlock, [2, 2, 2, 2], **kwargs)
    if pretrained:
        print('Loading the pretrained model ...')
        # strict = False as we don't need fc layer params.
        model.load_state_dict(model_zoo.load_url(model_urls['resnet18']), strict=False)
    return model

ResNet34

def resnet34(pretrained=False, **kwargs):
    """Constructs a ResNet-34 model.

    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet
    """
    model = ResNet(BasicBlock, [3, 4, 6, 3], **kwargs)
    if pretrained:
        print('Loading the pretrained model ...')
        model.load_state_dict(model_zoo.load_url(model_urls['resnet34']), strict=False)
    return model

ResNet20

這個(gè)需要強(qiáng)調(diào)一下,正常的ResNet20應(yīng)該是文章中提出，針對(duì)cifar數(shù)據(jù)集設(shè)計(jì)的n=3時(shí)候, 1+6*3+1=20

class ResNet4Cifar(nn.Module):
    def __init__(self, block, num_block, num_classes=10):
        super().__init__()
        self.in_channels = 16
        self.conv1 = nn.Sequential(
            nn.Conv2d(3, 16, kernel_size=3, padding=1, bias=False),
            nn.BatchNorm2d(16),
            nn.ReLU(inplace=True))
        # we use a different inputsize than the original paper
        # so conv2_x's stride is 1
        self.conv2_x = self._make_layer(block, 16, num_block[0], 1)
        self.conv3_x = self._make_layer(block, 32, num_block[1], 2)
        self.conv4_x = self._make_layer(block, 64, num_block[2], 2)
        self.avg_pool = nn.AdaptiveAvgPool2d((1, 1))
        self.fc = nn.Linear(64 * block.expansion, num_classes)

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

    def forward(self, x):
        output = self.conv1(x)
        output = self.conv2_x(output)
        output = self.conv3_x(output)
        output = self.conv4_x(output)
        output = self.avg_pool(output)
        output = output.view(output.size(0), -1)
        output = self.fc(output)
        return output

def resnet20(num_classes=10, **kargs):
    """ return a ResNet 20 object
    """
    return ResNet4Cifar(BasicBlock, [3, 3, 3], num_classes=num_classes)

我們通過(guò)參數(shù)量的計(jì)算也為0.27M，和論文中的一致，對(duì)[1,3,32,32]的輸入，輸出維度為[1,64,8,8]

但是也有一些文章只換了開頭三層的3x3卷積層，通道數(shù)并沒(méi)有采用16、32、64，仍是4層的64、128、256、512
，這樣下來(lái)參數(shù)量是11.25M。針對(duì)的任務(wù)不同，但是如果不關(guān)注原始網(wǎng)絡(luò)結(jié)構(gòu)，這一點(diǎn)可以忽略。

Bottleneck Block

Bottleneck Block中使用了1×1卷積層。如輸入通道數(shù)為256，1×1卷積層會(huì)將通道數(shù)先降為64，經(jīng)過(guò)3×3卷積層后，再將通道數(shù)升為256。1×1卷積層的優(yōu)勢(shì)是在更深的網(wǎng)絡(luò)中，用較小的參數(shù)量處理通道數(shù)很大的輸入。
這種結(jié)構(gòu)用在ResNet50、ResNet101中。

class Bottleneck(nn.Module):
    expansion = 4

    def __init__(self, inplanes, planes, stride=1, downsample=None):
        super(Bottleneck, self).__init__()
        self.conv1 = conv1x1(inplanes, planes)
        self.bn1 = nn.BatchNorm2d(planes)
        self.conv2 = conv3x3(planes, planes, stride)
        self.bn2 = nn.BatchNorm2d(planes)
        self.conv3 = conv1x1(planes, planes * self.expansion)
        self.bn3 = nn.BatchNorm2d(planes * self.expansion)
        self.relu = nn.ReLU(inplace=True)
        self.downsample = downsample
        self.stride = stride

    def forward(self, x):
        identity = x

        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

        out = self.conv2(out)
        out = self.bn2(out)
        out = self.relu(out)

        out = self.conv3(out)
        out = self.bn3(out)

        if self.downsample is not None:
            identity = self.downsample(x)

        out += identity
        out = self.relu(out)

        return out

ResNet50

和以上的網(wǎng)絡(luò)結(jié)構(gòu)一樣，把Bottleneck按層數(shù)堆起來(lái)就可以了

def resnet50(pretrained=False, **kwargs):
    """Constructs a ResNet-50 model.

    Args:
        pretrained (bool): If True, returns a model pre-trained on ImageNet
    """
    model = ResNet(Bottleneck, [3, 4, 6, 3], **kwargs)
    if pretrained:
        print('Loading the pretrained model ...')
        model.load_state_dict(model_zoo.load_url(model_urls['resnet50']), strict=False)
    return model

ResNet到底解決了什么問(wèn)題

推薦看知乎問(wèn)題Resnet到底在解決一個(gè)什么問(wèn)題呢？
貼一些我比較喜歡的回答：

A. 對(duì)于$L$層的網(wǎng)絡(luò)來(lái)說(shuō)，沒(méi)有殘差表示的Plain Net梯度相關(guān)性的衰減在$\frac{1}{2^L}$ ，而ResNet的衰減卻只有$\frac{1}{\sqrt{L}}$ 。即使BN過(guò)后梯度的模穩(wěn)定在了正常范圍內(nèi)，但梯度的相關(guān)性實(shí)際上是隨著層數(shù)增加持續(xù)衰減的。而經(jīng)過(guò)證明，ResNet可以有效減少這種相關(guān)性的衰減。

B. 對(duì)于“梯度彌散”觀點(diǎn)來(lái)說(shuō)，在輸出引入一個(gè)輸入x的恒等映射，則梯度也會(huì)對(duì)應(yīng)地引入一個(gè)常數(shù)1，這樣的網(wǎng)絡(luò)的確不容易出現(xiàn)梯度值異常，在某種意義上，起到了穩(wěn)定梯度的作用。

C. 跳連接相加可以實(shí)現(xiàn)不同分辨率特征的組合，因?yàn)闇\層容易有高分辨率但是低級(jí)語(yǔ)義的特征，而深層的特征有高級(jí)語(yǔ)義，但分辨率就很低了。引入跳接實(shí)際上讓模型自身有了更加“靈活”的結(jié)構(gòu)，即在訓(xùn)練過(guò)程本身，模型可以選擇在每一個(gè)部分是“更多進(jìn)行卷積與非線性變換”還是“更多傾向于什么都不做”，抑或是將兩者結(jié)合。模型在訓(xùn)練便可以自適應(yīng)本身的結(jié)構(gòu)。³

D. 當(dāng)使用了殘差網(wǎng)絡(luò)時(shí),就是加入了skip connection 結(jié)構(gòu),這時(shí)候由一個(gè)building block 的任務(wù)由: F(x) := H(x)，變成了F(x) := H(x)-x對(duì)比這兩個(gè)待擬合的函數(shù), 擬合殘差圖更容易優(yōu)化,也就是說(shuō):F(x) := H(x)-x比F(x) := H(x)更容易優(yōu)化⁴. 舉了一個(gè)差分放大器的例子：F是求和前網(wǎng)絡(luò)映射，H是從輸入到求和后的網(wǎng)絡(luò)映射。比如把5映射到5.1，那么引入殘差前是F’(5)=5.1，引入殘差后是H(5)=5.1, H(5)=F(5)+5, F(5)=0.1。這里的F’和F都表示網(wǎng)絡(luò)參數(shù)映射，引入殘差后的映射對(duì)輸出的變化更敏感。比如s輸出從5.1變到5.2，映射F’的輸出增加了1/51=2%，而對(duì)于殘差結(jié)構(gòu)輸出從5.1到5.2，映射F是從0.1到0.2，增加了100%。明顯后者輸出變化對(duì)權(quán)重的調(diào)整作用更大，所以效果更好。殘差的思想都是去掉相同的主體部分，從而突出微小的變化。

說(shuō)法眾多，好用就完事兒了嗷~

到了這里，關(guān)于【pytorch】ResNet18、ResNet20、ResNet34、ResNet50網(wǎng)絡(luò)結(jié)構(gòu)與實(shí)現(xiàn)的文章就介紹完了。如果您還想了解更多內(nèi)容，請(qǐng)?jiān)谟疑辖撬阉鱐OY模板網(wǎng)以前的文章或繼續(xù)瀏覽下面的相關(guān)文章，希望大家以后多多支持TOY模板網(wǎng)！