注意力机制在CNN中使用总结-阿里云开发者社区

摘要

1、通道注意力机制和空间注意力机制

2、SE-Net: Squeeze-and-Excitation Networks

SE模块的实现

SE的另一种实现方式

3、轻量模块ECANet（通道注意力超强改进）

4、Coordinate Attention

摘要

计算机视觉（computer vision）中的注意力机制（attention）的基本思想就是想让系统学会注意力——能够忽略无关信息而关注重点信息。

注意力机制按照关注的域来分:

空间域(spatial domain)

通道域(channel domain)

层域(layer domain)

混合域(mixed domain)

时间域(time domain)：还有另一种比较特殊的强注意力实现的注意力域，时间域(time domain)，但是因为强注意力是使用reinforcement learning来实现的，训练起来有所不同

1、通道注意力机制和空间注意力机制

Convolutional Block Attention Module (CBAM) 表示卷积模块的注意力机制模块。是一种结合了空间（spatial）和通道（channel）的注意力机制模块。相比于senet只关注通道（channel）的注意力机制可以取得更好的效果。

通道注意力：将输入的featuremap，分别经过基于width和height的global max pooling 和global average pooling，然后分别经过MLP。将MLP输出的特征进行基于elementwise的加和操作，再经过sigmoid激活操作，生成最终的channel attention featuremap。将该channel attention featuremap和input featuremap做elementwise乘法操作，生成Spatial attention模块需要的输入特征。

空间注意力：将Channel attention模块输出的特征图作为本模块的输入特征图。首先做一个基于channel的global max pooling 和global average pooling，然后将这2个结果基于channel 做concat操作。然后经过一个卷积操作，降维为1个channel。再经过sigmoid生成spatial attention feature。最后将该feature和该模块的输入feature做乘法，得到最终生成的特征。

代码如下：

import torch.nn as nn
import math
try:
    from torch.hub import load_state_dict_from_url
except ImportError:
    from torch.utils.model_zoo import load_url as load_state_dict_from_url
import torch
#通道注意力机制
class ChannelAttention(nn.Module):
    def __init__(self, in_planes, ratio=16):
        super(ChannelAttention, self).__init__()
        self.avg_pool = nn.AdaptiveAvgPool2d(1)
        self.max_pool = nn.AdaptiveMaxPool2d(1)
        self.fc1   = nn.Conv2d(in_planes, in_planes // 16, 1, bias=False)
        self.relu1 = nn.ReLU()
        self.fc2   = nn.Conv2d(in_planes // 16, in_planes, 1, bias=False)
        self.sigmoid = nn.Sigmoid()
    def forward(self, x):
        avg_out = self.fc2(self.relu1(self.fc1(self.avg_pool(x))))
        max_out = self.fc2(self.relu1(self.fc1(self.max_pool(x))))
        out = avg_out + max_out
        return self.sigmoid(out)
#空间注意力机制
class SpatialAttention(nn.Module):
    def __init__(self, kernel_size=7):
        super(SpatialAttention, self).__init__()
        assert kernel_size in (3, 7), 'kernel size must be 3 or 7'
        padding = 3 if kernel_size == 7 else 1
        self.conv1 = nn.Conv2d(2, 1, kernel_size, padding=padding, bias=False)
        self.sigmoid = nn.Sigmoid()
    def forward(self, x):
        avg_out = torch.mean(x, dim=1, keepdim=True)
        max_out, _ = torch.max(x, dim=1, keepdim=True)
        x = torch.cat([avg_out, max_out], dim=1)
        x = self.conv1(x)
        return self.sigmoid(x)

使用举例，在Resnet网络中添加注意力机制

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__()
        if norm_layer is None:
            norm_layer = nn.BatchNorm2d
        self._norm_layer = norm_layer
        self.inplanes = 64
        self.dilation = 1
        if replace_stride_with_dilation is None:
            # each element in the tuple indicates if we should replace
            # the 2x2 stride with a dilated convolution instead
            replace_stride_with_dilation = [False, False, False]
        if len(replace_stride_with_dilation) != 3:
            raise ValueError("replace_stride_with_dilation should be None "
                             "or a 3-element tuple, got {}".format(replace_stride_with_dilation))
        self.groups = groups
        self.base_width = width_per_group
        self.conv1 = nn.Conv2d(3, self.inplanes, kernel_size=7, stride=2, padding=3,
                               bias=False)
        self.bn1 = norm_layer(self.inplanes)
        self.relu = nn.ReLU(inplace=True)
        # 网络的第一层加入注意力机制
        self.ca = ChannelAttention(self.inplanes)
        self.sa = SpatialAttention()
        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,
                                       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])
        # 网络的卷积层的最后一层加入注意力机制
        self.ca1 = ChannelAttention(self.inplanes)
        self.sa1 = SpatialAttention()
        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.GroupNorm)):
                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, 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:
            downsample = nn.Sequential(
                conv1x1(self.inplanes, planes * block.expansion, stride),
                norm_layer(planes * block.expansion),
            )
        layers = []
        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(self, x):
        x = self.conv1(x)
        x = self.bn1(x)
        x = self.relu(x)
        x = self.ca(x) * x
        x = self.sa(x) * x
        x = self.maxpool(x)
        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)
        x = self.ca1(x) * x
        x = self.sa1(x) * x
        x = self.avgpool(x)
        x = x.reshape(x.size(0), -1)
        x = self.fc(x)
        return x

注意点：因为不能改变ResNet的网络结构，所以CBAM不能加在block里面，因为加进去网络结构发生了变化，所以不能用预训练参数。加在最后一层卷积和第一层卷积不改变网络，可以用预训练参数。

添加位置：

# 网络的第一层加入注意力机制

self.ca = ChannelAttention(self.inplanes)

self.sa = SpatialAttention()

和

# 网络的卷积层的最后一层加入注意力机制

self.ca1 = ChannelAttention(self.inplanes)

self.sa1 = SpatialAttention()

forWord部分代码

x = self.ca(x) * x

x = self.sa(x) * x

x = self.maxpool(x)

x = self.layer1(x)

x = self.layer2(x)

x = self.layer3(x)

x = self.layer4(x)

x = self.ca1(x) * x

x = self.sa1(x) * x

2、SE-Net: Squeeze-and-Excitation Networks

论文链接：https://arxiv.org/abs/1709.01507

代码地址：https://github.com/hujie-frank/SENet

PyTorch代码地址：https://github.com/miraclewkf/SENet-PyTorch

SE-Net赢得了最后一届ImageNet 2017竞赛分类任务的冠军，其基本原理是对于每个输出channel，预测一个常数权重，对每个channel加权一下。结构如下图：

第一步每个通道H*W个数全局平均池化得到一个标量，称之为Squeeze，然后两个FC得到01之间的一个权重值，对原始的每个HxW的每个元素乘以对应通道的权重，得到新的feature map，称之为Excitation。任意的原始网络结构，都可以通过这个Squeeze-Excitation的方式进行feature recalibration，如下图。

具体实现上就是一个Global Average Pooling-FC-ReLU-FC-Sigmoid，第一层的FC会把通道降下来，然后第二层FC再把通道升上去，得到和通道数相同的C个权重，每个权重用于给对应的一个通道进行加权。上图中的r就是缩减系数，实验确定选取16，可以得到较好的性能并且计算量相对较小。SENet的核心思想在于通过网络根据loss去学习特征权重，使得有效的feature map权重大，无效或效果小的feature map权重小的方式训练模型达到更好的结果。

SE模块的实现

这里给出PyTorch版本的实现（参考senet.pytorch）:

class SELayer(nn.Module):
    def __init__(self, channel, reduction=16):
        super(SELayer, self).__init__()
        self.avg_pool = nn.AdaptiveAvgPool2d(1)
        self.fc = nn.Sequential(
            nn.Linear(channel, channel // reduction, bias=False),
            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)
        y = self.fc(y).view(b, c, 1, 1)
        return x * y.expand_as(x)

将SE模块用在Resnet网络，只需要将SE模块加入到残差单元（应用在残差学习那一部分）就可以：

class SEBottleneck(nn.Module):
        expansion = 4
        def __init__(self, inplanes, planes, stride=1, downsample=None, reduction=16):
            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 = downsample
            self.stride = stride
        def forward(self, x):
            residual = 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)
            out = self.se(out)
            if self.downsample is not None:
                residual = self.downsample(x)
            out += residual
            out = self.relu(out)
            return out

SE的另一种实现方式

该方式使用卷积替代全连接。

class SEBlock(nn.Module):
    def __init__(self, input_channels, internal_neurons):
        super(SEBlock, self).__init__()
        self.down = nn.Conv2d(in_channels=input_channels, out_channels=internal_neurons, kernel_size=1, stride=1,
                              bias=True, padding_mode='same')
        self.up = nn.Conv2d(in_channels=internal_neurons, out_channels=input_channels, kernel_size=1, stride=1,
                            bias=True, padding_mode='same')
    def forward(self, inputs):
        x = F.avg_pool2d(inputs, kernel_size=inputs.size(3))
        x = self.down(x)
        x = F.leaky_relu(x)
        x = self.up(x)
        x = F.sigmoid(x)
        x = x.repeat(1, 1, inputs.size(2), inputs.size(3))
        return inputs * x

3、轻量模块ECANet（通道注意力超强改进）

论文链接：https://arxiv.org/abs/1910.03151

代码地址：https://github.com/BangguWu/ECANet

论文翻译：https://wanghao.blog.csdn.net/article/details/113073026

ECANet主要对SENet模块进行了一些改进，提出了一种不降维的局部跨信道交互策略（ECA模块）和自适应选择一维卷积核大小的方法，从而实现了性能上的提优。

ECANet的实现：

class eca_layer(nn.Module):
    """Constructs a ECA module.
    Args:
        channel: Number of channels of the input feature map
        k_size: Adaptive selection of kernel size
    """
    def __init__(self, channel, k_size=3):
        super(eca_layer, self).__init__()
        self.avg_pool = nn.AdaptiveAvgPool2d(1)
        self.conv = nn.Conv1d(1, 1, kernel_size=k_size, padding=(k_size - 1) // 2, bias=False) 
        self.sigmoid = nn.Sigmoid()
    def forward(self, x):
        # x: input features with shape [b, c, h, w]
        b, c, h, w = x.size()
        # feature descriptor on the global spatial information
        y = self.avg_pool(x)
        # Two different branches of ECA module
        y = self.conv(y.squeeze(-1).transpose(-1, -2)).transpose(-1, -2).unsqueeze(-1)
        # Multi-scale information fusion
        y = self.sigmoid(y)
        return x * y.expand_as(x)