resnet 1808部署 resnet网络搭建

文章目录

• 亮点
• 残差结构

• 计算量
• 虚线残差结构

• 代码解析

• resnet18/34的残差结构
• resnet50/101/152的残差结构Bottleneck
• 一层layer的结构(_make_layer()函数)
• ResNet主网络

• 代码仓库

亮点

• 引入了残差结构
• 使用Batch Normalization加速训练(丢弃dropout)

这两个方法，解决了梯度消失和梯度爆炸等问题，使得构建深层网络成为可能

残差结构

计算量

左边是ResNet18/34的残差结构，右边是ResNet101/152的残差结构

1. 左边计算量：3x3x256x256+3x3x256x256=1179648
2. 右边计算量：1x1x256x64+3x3x64x64+1x1x64x256=69632

由此可见，右边一个残差结构的计算量更少，原因是右边的残差结构，使用阿一个1*1的卷积核，用来降维和升维

虚线残差结构

上图有两个残差结构，它们的区别在于：

• 左图的输入(Input)直接和输出(Output)相加，而右图的输入(Input2)需要经过一个1*1的卷积核，才能与输出(Output2)相加。
• 虚线残差第一个卷积的步距stride=2,而实线残差结构stride=1

下图中，1号框的64表示卷积核的个数，等于输出深度。2号框的既是虚线残差结构

代码解析

resnet18/34的残差结构

BasicBlock如下图所示，它包括实线残差和虚线残差两种结构

# resnet18和resnet34的残差结构
class BasicBlock(nn.Module):
    # 卷积核个数改变的倍数，如果一样，则expansion=1
    expansion = 1

    # ----------------------------残差结构--------------------------------------------
    # in_channel        输入特征矩阵的深度
    # out_channel       输出特征矩阵的深度，既是卷积核的个数
    # stride            步距，当stride=1时，宽高不变，当stride=2时，宽高为原来的一半
    # downsample        下采样，默认为None。只有使用到虚线残差结构才设为True
    # bias=False        使用BN结构时，不需要使用偏置(bias)
    # -------------------------------------------------------------------------------
    def __init__(self, in_channel, out_channel, stride=1, downsample=None, **kwargs):
        super(BasicBlock, self).__init__()
        self.conv1 = nn.Conv2d(in_channels=in_channel, out_channels=out_channel,
                               kernel_size=3, stride=stride, padding=1, bias=False)
        self.bn1 = nn.BatchNorm2d(out_channel)
        self.relu = nn.ReLU()
        self.conv2 = nn.Conv2d(in_channels=out_channel, out_channels=out_channel,
                               kernel_size=3, stride=1, padding=1, bias=False)
        self.bn2 = nn.BatchNorm2d(out_channel)
        self.downsample = downsample

    def forward(self, x):
        identity = x
        # 判断self.downsample是否为空，如果不为空，则进行实线的残差结构
        if self.downsample is not None:
            identity = self.downsample(x)

        # 主支线的正向传播
        out = self.conv1(x)
        out = self.bn1(out)
        out = self.relu(out)

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

        # 主线和残差相加
        out += identity
        out = self.relu(out)

        return out

下面的代码段，对应的是一次卷积操作

self.conv1 = nn.Conv2d(in_channels=in_channel, out_channels=out_channel,kernel_size=3, stride=stride, padding=1, bias=False)
self.bn1 = nn.BatchNorm2d(out_channel)
self.relu = nn.ReLU()

重复两次，既有

# ----------------------------残差结构--------------------------------------------
    # in_channel        输入特征矩阵的深度
    # out_channel       输出特征矩阵的深度，既是卷积核的个数
    # stride            步距，当stride=1时，宽高不变，当stride=2时，宽高为原来的一半
    # downsample        下采样，默认为None。只有使用到虚线残差结构才设为True
    # bias=False        使用BN结构时，不需要使用偏置(bias)
    # -------------------------------------------------------------------------------
    def __init__(self, in_channel, out_channel, stride=1, downsample=None, **kwargs):
        super(BasicBlock, self).__init__()
        self.conv1 = nn.Conv2d(in_channels=in_channel, out_channels=out_channel,
                               kernel_size=3, stride=stride, padding=1, bias=False)
        self.bn1 = nn.BatchNorm2d(out_channel)
        self.relu = nn.ReLU()
        self.conv2 = nn.Conv2d(in_channels=out_channel, out_channels=out_channel,
                               kernel_size=3, stride=1, padding=1, bias=False)
        self.bn2 = nn.BatchNorm2d(out_channel)
        self.downsample = downsample

self.downsample = downsample对应的下采样，即是用一个1*1的卷积实现虚线残差结构。如果self.downsample为空，机型实线残差。如果不为空，则进行虚线的残差结构

out += identity进行残差结构+主线结构

resnet50/101/152的残差结构Bottleneck

class Bottleneck(nn.Module):
    # 卷积核个数改变的倍数，这里主干上的输出深度为输入深度的4倍，则expansion=4
    expansion = 4

    def __init__(self, in_channel, out_channel, stride=1, downsample=None,
                 groups=1, width_per_group=64):
        super(Bottleneck, self).__init__()

        width = int(out_channel * (width_per_group / 64.)) * groups

        self.conv1 = nn.Conv2d(in_channels=in_channel, out_channels=width,
                               kernel_size=1, stride=1, bias=False)  # squeeze channels
        self.bn1 = nn.BatchNorm2d(width)
        # -----------------------------------------
        self.conv2 = nn.Conv2d(in_channels=width, out_channels=width, groups=groups,
                               kernel_size=3, stride=stride, bias=False, padding=1)
        self.bn2 = nn.BatchNorm2d(width)
        # -----------------------------------------
        self.conv3 = nn.Conv2d(in_channels=width, out_channels=out_channel*self.expansion,
                               kernel_size=1, stride=1, bias=False)  # unsqueeze channels
        self.bn3 = nn.BatchNorm2d(out_channel*self.expansion)
        self.relu = nn.ReLU(inplace=True)
        self.downsample = downsample

Bottleneck与BasicBlock相似，仅在几个地方有些许差异

1. expansion = 4:conv1和conv3的输出深度分别为64、256，它们相差了4倍，即是expansion，它在self.conv3 = nn.Conv2d(in_channels=width, out_channels=out_channel*self.expansion,kernel_size=1, stride=1, bias=False)中被使用
2. self.conv2 = nn.Conv2d(in_channels=width, out_channels=width, groups=groups,kernel_size=3, stride=stride, bias=False, padding=1)：注意stride=stride，它对应着虚线残差结构中的conv2，stride=2

一层layer的结构(_make_layer()函数)

# ----------------------------一个layer的结构--------------------------------------
# block             残差结构,有BasicBlock、Bottleneck
# channel           与blocks_num对应,残差结构的卷积核数目，为一个列表。如resnet18为[2,2,2,2]
# block_num         该层一共包含了几个残差块，即使执行的次数
# -------------------------------------------------------------------------------
def _make_layer(self, block, channel, block_num, stride=1):
    downsample = None

    # 判断是resnet18/resnet34还是resnet50/resnet101/resnet152
    if stride != 1 or self.in_channel != channel * block.expansion:
        downsample = nn.Sequential(
            nn.Conv2d(self.in_channel, channel * block.expansion, kernel_size=1, stride=stride, bias=False),
            nn.BatchNorm2d(channel * block.expansion))

    # 若有downsample则为虚线残差结构，否则，仍是虚线残差结构
    layers = []
    layers.append(block(self.in_channel,
                        channel,
                        downsample=downsample,
                        stride=stride,
                        groups=self.groups,
                        width_per_group=self.width_per_group))
    self.in_channel = channel * block.expansion

    # 实线残差结构
    for _ in range(1, block_num):
        layers.append(block(self.in_channel,
                            channel,
                            groups=self.groups,
                            width_per_group=self.width_per_group))

    return nn.Sequential(*layers)

block：残差结构,有BasicBlock、Bottleneck

channel：与blocks_num对应,残差结构的卷积核数目，为一个列表。如resnet18为[2,2,2,2]

block_num：该层一共包含了几个残差块，即使执行的次数

1. downsample = None默认下采样为空，即默认不执行1*1卷积核的虚线残差结构
2. if stride != 1 or self.in_channel != channel * block.expansion:默认in_channel=64，channel=64；对于18、34来说：expansion=1，所以self.in_channel != channel * block.expansion成立，故不进入if语句。对于50、101来说：expansion=4，所以不相等，进入if语句
3. if语句就是一个1*1的卷积操作

下图的【💔红色】1对应着代码1部分；【💚绿色】2代表的代码2部分

ResNet主网络

class ResNet(nn.Module):

    def __init__(self,
                 block,
                 blocks_num,
                 num_classes=1000,
                 include_top=True,
                 groups=1,
                 width_per_group=64):
        super(ResNet, self).__init__()
        self.include_top = include_top
        self.in_channel = 64

        self.groups = groups
        self.width_per_group = width_per_group

        self.conv1 = nn.Conv2d(3, self.in_channel, kernel_size=7, stride=2,
                               padding=3, bias=False)
        self.bn1 = nn.BatchNorm2d(self.in_channel)
        self.relu = nn.ReLU(inplace=True)
        self.maxpool = nn.MaxPool2d(kernel_size=3, stride=2, padding=1)
        self.layer1 = self._make_layer(block, 64, blocks_num[0])
        self.layer2 = self._make_layer(block, 128, blocks_num[1], stride=2)
        self.layer3 = self._make_layer(block, 256, blocks_num[2], stride=2)
        self.layer4 = self._make_layer(block, 512, blocks_num[3], stride=2)
        if self.include_top:
            self.avgpool = nn.AdaptiveAvgPool2d((1, 1))  # output size = (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')