SE-ResNet的实现

见：D:\pythonCodes\深度学习实验\4.1_经典分类网络\inference代码汇总\models\se_resnet.py

一、SE-ResNet的实现方法

读了senet这篇论文之后，可以知道senet并没有提出一个新的网络，而是提出了一个即插即用的模块。这个模块叫做SE Block（在实现的时候，为了防止与SEBasicBlock这个名字混淆，叫做SELayer）。

本文希望实现se_resnet网络，也就是将SE Block嵌入到ResNet中形成的网络。se_resnet与resnet的差别就是，就是在BasicBlock（resnet18/34使用的是BasicBlock堆叠，而resnet50/101/152使用的是Bottleneck进行堆叠，这里就以BasicBlock举例，Bottleneck完全一样）中增加了SE Block这个操作。

比如下图，上面是BasicBlock的结构，下图就是SEBasic的结构，就是多出来了一个小圈圈。

通过读resnet的源码，我们知道是通过Resnet()这个类来组织成整个网络的。比如：

resnet34 = ResNet(BasicBlock, [3,4,6,3])

resnet18 = ResNet(BasicBlock, [2,2,2,2])

resnet50 = ResNet(Bottleneck, [3,4,6,3])

resnet101 = ResNet(Bottleneck, [3,4,23,3])

resnet152 = ResNet(Bottleneck, [3,8,36,3])

ResNet()接收两个参数，一个是block，另一个是堆叠的次数layers。只要传入参数，就能组织成一个网络了。比如传入的是BasicBlock，[3,4,6,3]就能得到resnet34了。这个函数就会自动地用3个BasicBlock组成layer1,用4个BasicBlock组成layer2,用6个BasicBlock组成layer3，用3个BasicBlock组成layer3，然后加上头尾等，组成一个网络。

我们可以利用ResNet()函数来构建我们的se_resent网络。只要给ResNet()传入SEBasicBlock和[3,4,6,3]就可以得到se_resnet34了。。

因此最关键的就是实现SEBasicBlock。而SEBasicBlock代码简直就是照抄BasicBlock代码，只要加上SELayer就行了。

(1) SELayer的实现

就是论文中的SE Block，在实现的时候，为了防止与SEBasicBlock这个名字混淆，叫做SELayer。

就是实现下面这个操作：

代码：

class SELayer(nn.Module):def __init__(self, channel, reduction=16):super(SELayer, self).__init__()self.avg_pool = nn.AdaptiveAvgPool2d(1)        #全局平均池化，输入BCHW -> 输出 B*C*1*1self.fc = nn.Sequential(nn.Linear(channel, channel // reduction, bias=False),   #可以看到channel得被reduction整除，否则可能出问题nn.ReLU(inplace=True),nn.Linear(channel // reduction, channel, bias=False),nn.Sigmoid())def forward(self, x):b, c, _, _ = x.size()y = self.avg_pool(x).view(b, c)     #得到B*C*1*1,然后转成B*C，才能送入到FC层中。y = self.fc(y).view(b, c, 1, 1)     #得到B*C的向量，C个值就表示C个通道的权重。把B*C变为B*C*1*1是为了与四维的x运算。return x * y.expand_as(x)           #先把B*C*1*1变成B*C*H*W大小，其中每个通道上的H*W个值都相等。*表示对应位置相乘。

（2）SEBasicBlock的写法

有了SELayer之后，就可以很容易地写SEBasicBlock了。就是照抄BasicBlock代码，只要加上SELayer就行了。

class SEBasicBlock(nn.Module):expansion = 1def __init__(self, inplanes, planes, stride=1, downsample=None, groups=1,base_width=64, dilation=1, norm_layer=None,*, reduction=16):# 参数列表里的 * 星号，标志着位置参数的就此终结，之后的那些参数，都只能以关键字形式来指定。super(SEBasicBlock, self).__init__()self.conv1 = conv3x3(inplanes, planes, stride)self.bn1 = nn.BatchNorm2d(planes)self.relu = nn.ReLU(inplace=True)self.conv2 = conv3x3(planes, planes, 1)self.bn2 = nn.BatchNorm2d(planes)self.se = SELayer(planes, reduction)self.downsample = downsampleself.stride = stridedef forward(self, x):residual = xout = self.conv1(x)out = self.bn1(out)out = self.relu(out)out = self.conv2(out)out = self.bn2(out)out = self.se(out)if self.downsample is not None:residual = self.downsample(x)out += residualout = self.relu(out)return out

就是多了以下两句：

其余的与senet中的basicBlock完全一致。SEBottleneck也是比Bottleneck多了这么一点。

二、完整代码

有了SELayer和SEBasicBlock之后就可以，借用resnet中的ResNet类来搭建SENet网络了。完整的代码如下，包含se_resnet18/34/50/101/152的实现。

# -*- coding: utf-8 -*-
"""
# @file name  : se_resnet.py
# @author     : https://github.com/moskomule/senet.pytorch
# @date       : 2020-08-07
# @brief      : se_resnet 模型搭建
"""
import torch
import torch.nn as nn
from torch.hub import load_state_dict_from_url
from torchvision.models import ResNet# 论文核心 SE Block, 这里称为 SE layer
class SELayer(nn.Module):def __init__(self, channel, reduction=16):super(SELayer, self).__init__()self.avg_pool = nn.AdaptiveAvgPool2d(1)        #全局平均池化，输入BCHW -> 输出 B*C*1*1self.fc = nn.Sequential(nn.Linear(channel, channel // reduction, bias=False), #可以看到channel得被reduction整除，否则可能出问题nn.ReLU(inplace=True),nn.Linear(channel // reduction, channel, bias=False),nn.Sigmoid())def forward(self, x):b, c, _, _ = x.size()y = self.avg_pool(x).view(b, c)   #得到B*C*1*1,然后转成B*C，才能送入到FC层中。y = self.fc(y).view(b, c, 1, 1)   #得到B*C的向量，C个值就表示C个通道的权重。把B*C变为B*C*1*1是为了与四维的x运算。return x * y.expand_as(x)         #先把B*C*1*1变成B*C*H*W大小，其中每个通道上的H*W个值都相等。*表示对应位置相乘。def conv3x3(in_planes, out_planes, stride=1):return nn.Conv2d(in_planes, out_planes, kernel_size=3, stride=stride, padding=1, bias=False)class SEBasicBlock(nn.Module):expansion = 1def __init__(self, inplanes, planes, stride=1, downsample=None, groups=1,base_width=64, dilation=1, norm_layer=None,*, reduction=16):# 参数列表里的 * 星号，标志着位置参数的就此终结，之后的那些参数，都只能以关键字形式来指定。super(SEBasicBlock, self).__init__()self.conv1 = conv3x3(inplanes, planes, stride)self.bn1 = nn.BatchNorm2d(planes)self.relu = nn.ReLU(inplace=True)self.conv2 = conv3x3(planes, planes, 1)self.bn2 = nn.BatchNorm2d(planes)self.se = SELayer(planes, reduction)self.downsample = downsampleself.stride = stridedef forward(self, x):residual = xout = self.conv1(x)out = self.bn1(out)out = self.relu(out)out = self.conv2(out)out = self.bn2(out)out = self.se(out)if self.downsample is not None:residual = self.downsample(x)out += residualout = self.relu(out)return outclass SEBottleneck(nn.Module):expansion = 4def __init__(self, inplanes, planes, stride=1, downsample=None, groups=1,base_width=64, dilation=1, norm_layer=None,*, reduction=16):# 参数列表里的 * 星号，标志着位置参数的就此终结，之后的那些参数，都只能以关键字形式来指定。super(SEBottleneck, self).__init__()self.conv1 = nn.Conv2d(inplanes, planes, kernel_size=1, bias=False)self.bn1 = nn.BatchNorm2d(planes)self.conv2 = nn.Conv2d(planes, planes, kernel_size=3, stride=stride,padding=1, bias=False)self.bn2 = nn.BatchNorm2d(planes)self.conv3 = nn.Conv2d(planes, planes * 4, kernel_size=1, bias=False)self.bn3 = nn.BatchNorm2d(planes * 4)self.relu = nn.ReLU(inplace=True)self.se = SELayer(planes * 4, reduction)self.downsample = downsampleself.stride = stridedef forward(self, x):residual = xout = 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)out = self.se(out)if self.downsample is not None:residual = self.downsample(x)out += residualout = self.relu(out)return outdef se_resnet18(num_classes=1_000):model = ResNet(SEBasicBlock, [2, 2, 2, 2], num_classes=num_classes)model.avgpool = nn.AdaptiveAvgPool2d(1)return modeldef se_resnet34(num_classes=1_000):model = ResNet(SEBasicBlock, [3, 4, 6, 3], num_classes=num_classes)model.avgpool = nn.AdaptiveAvgPool2d(1)return modeldef se_resnet50(num_classes=1_000, pretrained=False):model = ResNet(SEBottleneck, [3, 4, 6, 3], num_classes=num_classes)model.avgpool = nn.AdaptiveAvgPool2d(1)if pretrained:model.load_state_dict(load_state_dict_from_url("https://github.com/moskomule/senet.pytorch/releases/download/archive/seresnet50-60a8950a85b2b.pkl"))return modeldef se_resnet101(num_classes=1_000):model = ResNet(SEBottleneck, [3, 4, 23, 3], num_classes=num_classes)model.avgpool = nn.AdaptiveAvgPool2d(1)return modeldef se_resnet152(num_classes=1_000):model = ResNet(SEBottleneck, [3, 8, 36, 3], num_classes=num_classes)model.avgpool = nn.AdaptiveAvgPool2d(1)return modelif __name__ == "__main__":inputs = torch.randn(2, 3, 224, 224)model = se_resnet50(pretrained=False)outputs = model(inputs)print(outputs.size())

运行结果：