YOLOv5-7.0改进（四）添加EMA注意力机制

前言

关于网络中注意力机制的改进有很多种，本篇内容从EMA注意力机制开始！

往期回顾

YOLOv5-7.0改进（一）MobileNetv3替换主干网络

YOLOv5-7.0改进（二）BiFPN替换Neck网络

YOLOv5-7.0改进（三）添加损失函数EIoU、AlphaIoU、SIoU、WIoU、MPDIoU、NWD

一、EMA简介

论文题目：Efficient Multi-Scale Attention Module with Cross-Spatial Learning

EMA注意力机制：基于跨空间学习的高效多尺度注意力机制，该模块首先将部分通道维度重塑为批量维度，以避免通用卷积进行某种形式的降维，接着在每个并行子网络中构建局部的跨通道交互，利用一种新的跨空间学习方法融合两个并行子网络的输出特征图，设计了一个多尺度并行子网络来建立长短依赖关系。

网络结构：

二、Neck端添加EMA

第一步：在common.py中添加EMA模块

代码如下：

#EMA
class EMA(nn.Module):def __init__(self, channels, factor=8):super(EMA, self).__init__()self.groups = factor # 分组因子assert channels // self.groups > 0self.softmax = nn.Softmax(-1) #softmax操作self.agp = nn.AdaptiveAvgPool2d((1, 1)) # 1×1平均池化层self.pool_h = nn.AdaptiveAvgPool2d((None, 1)) # X平均池化层 h=1self.pool_w = nn.AdaptiveAvgPool2d((1, None)) # Y平均池化层 w=1self.gn = nn.GroupNorm(channels // self.groups, channels // self.groups) # 分组操作self.conv1x1 = nn.Conv2d(channels // self.groups, channels // self.groups, kernel_size=1, stride=1, padding=0) # 1×1卷积分支 self.conv3x3 = nn.Conv2d(channels // self.groups, channels // self.groups, kernel_size=3, stride=1, padding=1) # 3×3卷积分支def forward(self, x):b, c, h, w = x.size()group_x = x.reshape(b * self.groups, -1, h, w)  # b*g,c//g,h,wx_h = self.pool_h(group_x) # 得到平均池化之后的hx_w = self.pool_w(group_x).permute(0, 1, 3, 2) # 得到平均池化之后的whw = self.conv1x1(torch.cat([x_h, x_w], dim=2)) # 先拼接，然后送入1×1卷积x_h, x_w = torch.split(hw, [h, w], dim=2)x1 = self.gn(group_x * x_h.sigmoid() * x_w.permute(0, 1, 3, 2).sigmoid())x2 = self.conv3x3(group_x) # 3×3卷积分支x11 = self.softmax(self.agp(x1).reshape(b * self.groups, -1, 1).permute(0, 2, 1))x12 = x2.reshape(b * self.groups, c // self.groups, -1)  # b*g, c//g, hwx21 = self.softmax(self.agp(x2).reshape(b * self.groups, -1, 1).permute(0, 2, 1))x22 = x1.reshape(b * self.groups, c // self.groups, -1)  # b*g, c//g, hwweights = (torch.matmul(x11, x12) + torch.matmul(x21, x22)).reshape(b * self.groups, 1, h, w)return (group_x * weights.sigmoid()).reshape(b, c, h, w)

插入效果：

第二步：在yolo.py中的parse_model函数加入类名

将EMA类名添加到注册表中，效果如下：

第三步：制作模型配置文件

1、复制models/yolov5s.yaml文件，并重命名

2、将以下代码复制到新创建的yaml文件

# YOLOv5 🚀 by Ultralytics, GPL-3.0 license# Parameters
nc: 12  # number of classes
depth_multiple: 0.33  # model depth multiple
width_multiple: 0.50  # layer channel multiple
anchors:- [10,13, 16,30, 33,23]  # P3/8- [30,61, 62,45, 59,119]  # P4/16- [116,90, 156,198, 373,326]  # P5/32# YOLOv5 v6.0 backbone
backbone:# [from, number, module, args][[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2[-1, 1, Conv, [128, 3, 2]],  # 1-P2/4[-1, 3, C3, [128]],[-1, 1, Conv, [256, 3, 2]],  # 3-P3/8[-1, 6, C3, [256]],[-1, 1, Conv, [512, 3, 2]],  # 5-P4/16[-1, 9, C3, [512]],[-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32[-1, 3, C3, [1024]],[-1, 1, SPPF, [1024, 5]],  # 9]# YOLOv5 v6.0 head
head:[[-1, 1, Conv, [512, 1, 1]],[-1, 1, nn.Upsample, [None, 2, 'nearest']],[[-1, 6], 1, Concat, [1]],  # cat backbone P4[-1, 3, C3, [512, False]],  # 13[-1, 1, Conv, [256, 1, 1]],[-1, 1, nn.Upsample, [None, 2, 'nearest']],[[-1, 4], 1, Concat, [1]],  # cat backbone P3[-1, 3, C3, [256, False]],  # 17 (P3/8-small)[-1, 1, EMA, [256]],  # 加入到小目标层后[-1, 1, Conv, [256, 3, 2]],[[-1, 14], 1, Concat, [1]],  # cat head P4[-1, 3, C3, [512, False]],  # 20 (P4/16-medium)[-1, 1, EMA, [512]],  # 加入到中目标层后[-1, 1, Conv, [512, 3, 2]],[[-1, 10], 1, Concat, [1]],  # cat head P5[-1, 3, C3, [1024, False]],  # 23 (P5/32-large)[-1, 1, EMA, [1024]],  # 加入到大目标层后[[18, 22, 26], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)]

第四步：验证新加入的Neck网络

1、修改yolo.py中以下两个地方

（1）DetectionModel函数下的cfg

（2）parser = argparse.ArgumentParser()下的cfg

2、运行yolo.py

(1)yolov5s_EMA.yaml

好了，到这一步在Neck端添加EMA基本完成，接下就可以开始训练~

三、C3中添加EMA

第一步：在common.py中添加EMA模块

代码如下：

#EMA
class EMA(nn.Module):def __init__(self, channels, factor=8):super(EMA, self).__init__()self.groups = factor # 分组率assert channels // self.groups > 0self.softmax = nn.Softmax(-1) # Softmaxself.agp = nn.AdaptiveAvgPool2d((1, 1)) # 平均池化层self.pool_h = nn.AdaptiveAvgPool2d((None, 1)) # x平均池化层 h=1self.pool_w = nn.AdaptiveAvgPool2d((1, None)) # y平均池化层 w=1self.gn = nn.GroupNorm(channels // self.groups, channels // self.groups) # 分组操作self.conv1x1 = nn.Conv2d(channels // self.groups, channels // self.groups, kernel_size=1, stride=1, padding=0) # 1×1卷积分支self.conv3x3 = nn.Conv2d(channels // self.groups, channels // self.groups, kernel_size=3, stride=1, padding=1) # 3×3卷积分支def forward(self, x):b, c, h, w = x.size()group_x = x.reshape(b * self.groups, -1, h, w)  # b*g,c//g,h,wx_h = self.pool_h(group_x)x_w = self.pool_w(group_x).permute(0, 1, 3, 2)hw = self.conv1x1(torch.cat([x_h, x_w], dim=2))x_h, x_w = torch.split(hw, [h, w], dim=2)x1 = self.gn(group_x * x_h.sigmoid() * x_w.permute(0, 1, 3, 2).sigmoid())x2 = self.conv3x3(group_x)x11 = self.softmax(self.agp(x1).reshape(b * self.groups, -1, 1).permute(0, 2, 1))x12 = x2.reshape(b * self.groups, c // self.groups, -1)  # b*g, c//g, hwx21 = self.softmax(self.agp(x2).reshape(b * self.groups, -1, 1).permute(0, 2, 1))x22 = x1.reshape(b * self.groups, c // self.groups, -1)  # b*g, c//g, hwweights = (torch.matmul(x11, x12) + torch.matmul(x21, x22)).reshape(b * self.groups, 1, h, w)return (group_x * weights.sigmoid()).reshape(b, c, h, w)class C3_EMA3(nn.Module):# CSP Bottleneck with 3 convolutionsdef __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, number, shortcut, groups, expansionsuper().__init__()c_ = int(c2 * e)  # hidden channelsself.cv1 = Conv(c1, c_, 1, 1)self.cv2 = Conv(c1, c_, 1, 1)self.cv3 = Conv(2 * c_, c2, 1)  # optional act=FReLU(c2)self.m = nn.Sequential(*(Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)))self.m1 = nn.ModuleList([EMA(2 * c_)])  # 添加在最后一个卷积之前def forward(self, x):return self.cv3(self.m1[0](torch.cat((self.m(self.cv1(x)), self.cv2(x)), 1)))class C3_EMA2(nn.Module):# CSP Bottleneck with 3 convolutionsdef __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, number, shortcut, groups, expansionsuper().__init__()c_ = int(c2 * e)  # hidden channelsself.cv1 = Conv(c1, c_, 1, 1)self.cv2 = Conv(c1, c_, 1, 1)self.cv3 = Conv(2 * c_, c2, 1)  # optional act=FReLU(c2)self.m = nn.Sequential(*(Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)))self.m1 = nn.ModuleList([EMA(c1)])  # 添加在最后一个卷积之前def forward(self, x):return self.cv3(torch.cat((self.m(self.cv1(x)), self.cv2(self.m1[0](x))), 1))class C3_EMA1(nn.Module):# CSP Bottleneck with 3 convolutionsdef __init__(self, c1, c2, n=1, shortcut=True, g=1, e=0.5):  # ch_in, ch_out, number, shortcut, groups, expansionsuper().__init__()c_ = int(c2 * e)  # hidden channelsself.cv1 = Conv(c1, c_, 1, 1)self.cv2 = Conv(c1, c_, 1, 1)self.cv3 = Conv(2 * c_, c2, 1)  # optional act=FReLU(c2)self.m = nn.Sequential(*(Bottleneck(c_, c_, shortcut, g, e=1.0) for _ in range(n)))self.m1 = nn.ModuleList([EMA(c_)])  # 添加在最后一个卷积之前def forward(self, x):return self.cv3(torch.cat((self.m(self.m1[0](self.cv1(x))), self.cv2(x)), 1))

效果如下：

第二步：在yolo.py中的parse_model函数加入类名

将以下类名添加到注册表中

EMA, C3_EMA1, C3_EMA2, C3_EMA3

效果如下：

第三步：制作模型配置文件

将以下代码复制到yaml文件中

# YOLOv5 🚀 by Ultralytics, GPL-3.0 license# Parameters
nc: 12  # number of classes
depth_multiple: 0.33  # model depth multiple
width_multiple: 0.50  # layer channel multiple
anchors:- [10,13, 16,30, 33,23]  # P3/8- [30,61, 62,45, 59,119]  # P4/16- [116,90, 156,198, 373,326]  # P5/32# YOLOv5 v6.0 backbone
backbone:# [from, number, module, args][[-1, 1, Conv, [64, 6, 2, 2]],  # 0-P1/2[-1, 1, Conv, [128, 3, 2]],  # 1-P2/4[-1, 3, C3_EMA1, [128]],[-1, 1, Conv, [256, 3, 2]],  # 3-P3/8[-1, 6, C3, [256]],[-1, 1, Conv, [512, 3, 2]],  # 5-P4/16[-1, 9, C3, [512]],[-1, 1, Conv, [1024, 3, 2]],  # 7-P5/32[-1, 3, C3, [1024]],[-1, 1, SPPF, [1024, 5]],  # 9]# YOLOv5 v6.0 head
head:[[-1, 1, Conv, [512, 1, 1]],[-1, 1, nn.Upsample, [None, 2, 'nearest']],[[-1, 6], 1, Concat, [1]],  # cat backbone P4[-1, 3, C3, [512, False]],  # 13[-1, 1, Conv, [256, 1, 1]],[-1, 1, nn.Upsample, [None, 2, 'nearest']],[[-1, 4], 1, Concat, [1]],  # cat backbone P3[-1, 3, C3, [256, False]],  # 17 (P3/8-small)[-1, 1, Conv, [256, 3, 2]],[[-1, 14], 1, Concat, [1]],  # cat head P4[-1, 3, C3, [512, False]],  # 20 (P4/16-medium)[-1, 1, Conv, [512, 3, 2]],[[-1, 10], 1, Concat, [1]],  # cat head P5[-1, 3, C3, [1024, False]],  # 23 (P5/32-large)[[17, 20, 23], 1, Detect, [nc, anchors]],  # Detect(P3, P4, P5)]

第四步：验证新加入的Neck网络

1、运行yolo.py

接下来也是对这个模型进行训练，需要注意的是这是在主干网络部分改进~