优质广告供应商

广告是为了更好地支持作者创作

resnet18 50网络结构以及pytorch实现代码

1 resnet简介

  关于resnet,网上有大量的文章讲解其原理和思路,简单来说,resnet巧妙地利用了shortcut连接,解决了深度网络中模型退化的问题。

2 论文中的结构如下

网络结构.png

2.1 参考pytorch中的实现,自己画了一个网络图,包含了每一层的参数和输出

resnet18&resnet50.jpg

PS:经评论区@字里行间_yan提醒,原始图片中部分描述有歧义,已更正。一般来说,特征图的尺寸变化应表述为上采样和下采样,通道数的变化才是升维和降维。


2020/4/17更新
本来自己随便写写,感觉看到这篇文章的人挺多,回来填坑

  • 增加了pytorch中的代码解读。
  • 修复了图中参数k的标识错误(1x1卷积 k=1)
  • 新增SVG文件下载地址链接(虽然文件简单,下载后还请在本文点个赞鼓励下): 链接: 链接: https://pan.baidu.com/s/183ReRQMJXt2yUkhExnezBA 提取码: kmf2

3 pytorch中的resnet

3.1 代码文件

完整代码文件可在pytorch官方文档查到,地址https://pytorch.org/docs/stable/_modules/torchvision/models/resnet.html#resnet18
顺便说一句,torchvision里面实现了大部分经典的网络,分类分割检测都有,还包含了常用的一些数据库加载等,对于刚入门的同学来说会省很多事。

3.2 代码阅读

个人习惯从模型调用开始看,首先看调用

def _resnet(arch, block, layers, pretrained, progress, **kwargs):
    model = ResNet(block, layers, **kwargs)
    if pretrained:
        state_dict = load_state_dict_from_url(model_urls[arch],
                                              progress=progress)
        model.load_state_dict(state_dict)
    return model
def resnet50(pretrained=False, progress=True, **kwargs):
    return _resnet('resnet50', Bottleneck, [3, 4, 6, 3], pretrained, progress,
                   **kwargs)

  可以看到resnet至少需要两个显示的参数,分别是block和layers。这里的block就是论文里提到的resnet18和resnet50中应用的两种不同结构。layers就是网络层数,也就是每个block的个数,在前文图中也有体现。
  然后看网络结构,代码略长,为了阅读体验就直接截取了重要部分以及在代码中注释,建议配合完整代码阅读。

class ResNet(nn.Module):
    def __init__(self, block, layers, num_classes=1000, zero_init_residual=False,
                 groups=1, width_per_group=64, replace_stride_with_dilation=None,
                 norm_layer=None):
        super(ResNet, self).__init__()
        #参数比调用多几个,模型相较于最初发文章的时候有过更新
        #block: basicblock或者bottleneck,后续会提到
        #layers:每个block的个数,如resnet50, layers=[3,4,6,3]
        #num_classes: 数据库类别数量
        #zero_init_residual:其他论文中提到的一点小trick,残差参数为0
        #groups:卷积层分组,应该是为了resnext扩展
        #width_per_group:同上,此外还可以是wideresnet扩展
        #replace_stride_with_dilation:空洞卷积,非原论文内容
        #norm_layer:原论文用BN,此处设为可自定义
       
         # 中间部分代码省略,只看模型搭建部分
        self.layer1 = self._make_layer(block, 64, layers[0])
        self.layer2 = self._make_layer(block, 128, layers[1], stride=2,
                                       dilate=replace_stride_with_dilation[0])
        self.layer3 = self._make_layer(block, 256, layers[2], stride=2,
                                       dilate=replace_stride_with_dilation[1])
        self.layer4 = self._make_layer(block, 512, layers[3], stride=2,
                                       dilate=replace_stride_with_dilation[2])
        self.avgpool = nn.AdaptiveAvgPool2d((1, 1))
        self.fc = nn.Linear(512 * block.expansion, num_classes)

    #中间部分代码省略
    def _make_layer(self, block, planes, blocks, stride=1, dilate=False):
        norm_layer = self._norm_layer
        downsample = None
        previous_dilation = self.dilation
        if dilate:
            self.dilation *= stride
            stride = 1
        if stride != 1 or self.inplanes != planes * block.expansion:
           #当需要特征图需要降维或通道数不匹配的时候调用
            downsample = nn.Sequential(
                conv1x1(self.inplanes, planes * block.expansion, stride),
                norm_layer(planes * block.expansion),
            )

        layers = []
        #每一个self.layer的第一层需要调用downsample,所以单独写,跟下面range中的1 相对应
        #block的定义看下文
        layers.append(block(self.inplanes, planes, stride, downsample, self.groups,
                            self.base_width, previous_dilation, norm_layer))
        self.inplanes = planes * block.expansion
        for _ in range(1, blocks):
            layers.append(block(self.inplanes, planes, groups=self.groups,
                                base_width=self.base_width, dilation=self.dilation,
                                norm_layer=norm_layer))

        return nn.Sequential(*layers)

    def _forward_impl(self, x):
        #前向传播
        # See note [TorchScript super()]
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        x = self.maxpool(x)

        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)

        x = self.avgpool(x)
        x = torch.flatten(x, 1)
        x = self.fc(x)

        return x

然后是论文中的block

def conv3x3(in_planes, out_planes, stride=1, groups=1, dilation=1):
    """3x3 convolution with padding"""
    return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride,
                     padding=dilation, groups=groups, bias=False, dilation=dilation)


def conv1x1(in_planes, out_planes, stride=1):
    """1x1 convolution"""
    return nn.Conv2d(in_planes, out_planes, kernel_size=1, stride=stride, bias=False)

#用在resnet18中的结构,也就是两个3x3卷积
class BasicBlock(nn.Module):
    expansion = 1
    __constants__ = ['downsample']
    #inplanes:输入通道数
    #planes:输出通道数
    #base_width,dilation,norm_layer不在本文讨论范围
    def __init__(self, inplanes, planes, stride=1, downsample=None, groups=1,
                 base_width=64, dilation=1, norm_layer=None):
        super(BasicBlock, self).__init__()
        #中间部分省略
        self.conv1 = conv3x3(inplanes, planes, stride)
        self.bn1 = norm_layer(planes)
        self.relu = nn.ReLU(inplace=True)
        self.conv2 = conv3x3(planes, planes)
        self.bn2 = norm_layer(planes)
        self.downsample = downsample
        self.stride = stride

    def forward(self, x):
        #为后续相加保存输入
        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:
            #遇到降尺寸或者升维的时候要保证能够相加
            identity = self.downsample(x)

        out += identity#论文中最核心的部分,resnet的简洁和优美的体现
        out = self.relu(out)

        return out

#bottleneck是应用在resnet50及其以上的结构,主要是1x1,3x3,1x1
class Bottleneck(nn.Module):
    expansion = 4
    __constants__ = ['downsample']

    def __init__(self, inplanes, planes, stride=1, downsample=None, groups=1,
                 base_width=64, dilation=1, norm_layer=None):
        super(Bottleneck, self).__init__()
        #中间省略
        self.conv1 = conv1x1(inplanes, width)
        self.bn1 = norm_layer(width)
        self.conv2 = conv3x3(width, width, stride, groups, dilation)
        self.bn2 = norm_layer(width)
        self.conv3 = conv1x1(width, planes * self.expansion)
        self.bn3 = norm_layer(planes * self.expansion)
        self.relu = nn.ReLU(inplace=True)
        self.downsample = downsample
        self.stride = stride
    #同basicblock
    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

   我已经很熟悉resnet,不知道在哪些地方存在阅读问题,所以只在代码中把一些关键的地方注释了,代码可以跟我画的那个图结合起来看,更容易理解。
  如有问题欢迎留言,我看到就会回复。

优质广告供应商

广告是为了更好地支持作者创作

推荐阅读更多精彩内容