企业官网建设流程全解析

layui响应式网站开发教程,wordpress上传顶部图像,wordpress的自定义菜单图标,酒店网站建设栏目分析前言 本文介绍了将SENetV2与YOLOv8结合的方法#xff0c;以提升图像分类性能。SENetV2是结合Squeeze-and-Excitation#xff08;SE#xff09;模块和密集层的图像分类模型#xff0c;引入聚合稠密层用于通道和全局表示。其SE模块重新校准通道特征#xff0c;密集层优化特…前言本文介绍了将SENetV2与YOLOv8结合的方法以提升图像分类性能。SENetV2是结合Squeeze-and-ExcitationSE模块和密集层的图像分类模型引入聚合稠密层用于通道和全局表示。其SE模块重新校准通道特征密集层优化特征表示还提出SaE模块增强关键特征捕获。我们将SENetV2的SaELayer集成进YOLOv8在相关位置嵌入该模块。实验表明结合SENetV2的YOLOv8在图像分类准确性上有显著提升。文章目录 YOLOv8改进大全卷积层、轻量化、注意力机制、损失函数、Backbone、SPPF、Neck、检测头全方位优化汇总专栏链接: YOLOv8改进专栏文章目录前言介绍摘要文章链接基本原理SENetV2的结构如下SaE模块核心代码引入代码注册步骤1:步骤2配置yolov8_SENetV2.yaml实验脚本结果介绍摘要卷积神经网络CNNs通过提取空间特征彻底改变了图像分类并在基于视觉的任务中实现了最先进的准确性。提出的Squeeze-and-Excitation网络模块收集输入的通道表示。多层感知器MLP从数据中学习全局表示并在大多数图像分类模型中用于学习图像的提取特征。本文中我们引入了一种新型的聚合多层感知器一个多分支密集层嵌入到Squeeze-and-Excitation残差模块中旨在超越现有架构的性能。我们的方法结合了Squeeze-and-Excitation网络模块和密集层。这种融合增强了网络捕捉通道模式和全局知识的能力从而提高了特征表示。与SENet相比所提出的模型参数增加可以忽略不计。我们在基准数据集上进行了广泛的实验以验证模型并与已建立的架构进行比较。实验结果表明所提出模型在分类准确性上有显著提高。文章链接论文地址论文地址代码地址代码地址参考代码代码地址基本原理SENetV2是一种图像分类模型其核心特征是引入了聚合稠密层Aggregated Dense Layer用于通道和全局表示,是一种结合了Squeeze-and-ExcitationSE模块和密集层的图像分类模型。该模型旨在通过增强特征表示来提高图像分类性能。SENet V2的核心思想是通过对通道特征和全局特征进行重新校准和激活从而使网络更加专注于关键特征提高分类准确性。SENet V2的关键特点包括Squeeze-and-ExcitationSE模块SE模块通过对通道特征进行重新校准使网络能够更好地捕获关键特征。在SE模块中通过全局信息来动态调整通道特征的重要性从而提高网络的表达能力。密集层SENet V2引入了密集层用于进一步优化特征表示。密集层有助于增强通道特征的全局表示能力从而提高网络的分类性能。Squeeze Aggregated ExcitationSaE模块SENet V2还提出了SaE模块将聚合的密集层与SE模块相结合进一步优化特征表示。SaE模块通过增加层间的基数来优化关键特征的传输提高网络的性能。实验结果SENet V2在多个数据集上进行了实验评估包括CIFAR-10、CIFAR-100和ImageNet。实验结果表明SENet V2相较于传统架构在图像分类任务中取得了更高的准确性。SENetV2的结构如下SaE模块核心代码importtorchimporttorch.nnasnnimporttorch.nn.functionalasFfromtorch.utils.model_zooimportload_url# 定义 SE 模块classSELayer(nn.Module):def__init__(self,channel,reduction16):super(SELayer,self).__init__()# 全局平均池化层self.avg_poolnn.AdaptiveAvgPool2d(1)# 全连接层包含两层线性变换和激活函数self.fcnn.Sequential(nn.Linear(channel,channel//reduction,biasFalse),nn.ReLU(inplaceTrue),nn.Linear(channel//reduction,channel,biasFalse),nn.Sigmoid())defforward(self,x):b,c,_,_x.size()# 获取输入的维度yself.avg_pool(x).view(b,c)# 全局平均池化并改变维度yself.fc(y).view(b,c,1,1)# 通过全连接层并改变维度returnx*y.expand_as(x)# 按通道加权输入# 定义 SaE 模块classSaELayer(nn.Module):def__init__(self,in_channel,reduction32):super(SaELayer,self).__init__()# 检查输入通道数是否满足条件assertin_channelreductionandin_channel%reduction0,invalid in_channel in SaElayerself.reductionreduction self.cardinality4# 全局平均池化层self.avg_poolnn.AdaptiveAvgPool2d(1)# cardinality 1self.fc1nn.Sequential(nn.Linear(in_channel,in_channel//self.reduction,biasFalse),nn.ReLU(inplaceTrue))# cardinality 2self.fc2nn.Sequential(nn.Linear(in_channel,in_channel//self.reduction,biasFalse),nn.ReLU(inplaceTrue))# cardinality 3self.fc3nn.Sequential(nn.Linear(in_channel,in_channel//self.reduction,biasFalse),nn.ReLU(inplaceTrue))# cardinality 4self.fc4nn.Sequential(nn.Linear(in_channel,in_channel//self.reduction,biasFalse),nn.ReLU(inplaceTrue))# 最终的全连接层self.fcnn.Sequential(nn.Linear(in_channel//self.reduction*self.cardinality,in_channel,biasFalse),nn.Sigmoid())defforward(self,x):b,c,_,_x.size()# 获取输入的维度yself.avg_pool(x).view(b,c)# 全局平均池化并改变维度# 分别通过4个全连接层y1self.fc1(y)y2self.fc2(y)y3self.fc3(y)y4self.fc4(y)# 将4个输出拼接在一起y_concatetorch.cat([y1,y2,y3,y4],dim1)# 最终通过全连接层并改变维度y_ex_dimself.fc(y_concate).view(b,c,1,1)returnx*y_ex_dim.expand_as(x)# 按通道加权输入# 示例代码用于测试 SaELayer 模块se_v2SaELayer(64)# 示例输入inputtorch.randn(3,64,224,224)# 前向传播获取输出outputse_v2(input)# 打印输出的形状print(output.shape)# torch.Size([3, 64, 224, 224])引入代码在根目录下的ultralytics/nn/目录新建一个attention目录然后新建一个以SENetV2为文件名的py文件 把代码拷贝进去。importtorchimporttorch.nnasnnimporttorch.nn.functionalasFclassSELayer(nn.Module):def__init__(self,channel,reduction16):super(SELayer,self).__init__()self.avg_poolnn.AdaptiveAvgPool2d(1)self.fcnn.Sequential(nn.Linear(channel,channel//reduction,biasFalse),nn.ReLU(inplaceTrue),nn.Linear(channel//reduction,channel,biasFalse),nn.Sigmoid(),)defforward(self,x):b,c,_,_x.size()yself.avg_pool(x).view(b,c)yself.fc(y).view(b,c,1,1)returnx*y.expand_as(x)classSaELayer(nn.Module):def__init__(self,in_channel,reduction32):super(SaELayer,self).__init__()assert(in_channelreductionandin_channel%reduction0),invalid in_channel in SaElayerself.reductionreduction self.cardinality4self.avg_poolnn.AdaptiveAvgPool2d(1)# cardinality 1self.fc1nn.Sequential(nn.Linear(in_channel,in_channel//self.reduction,biasFalse),nn.ReLU(inplaceTrue),)# cardinality 2self.fc2nn.Sequential(nn.Linear(in_channel,in_channel//self.reduction,biasFalse),nn.ReLU(inplaceTrue),)# cardinality 3self.fc3nn.Sequential(nn.Linear(in_channel,in_channel//self.reduction,biasFalse),nn.ReLU(inplaceTrue),)# cardinality 4self.fc4nn.Sequential(nn.Linear(in_channel,in_channel//self.reduction,biasFalse),nn.ReLU(inplaceTrue),)self.fcnn.Sequential(nn.Linear(in_channel//self.reduction*self.cardinality,in_channel,biasFalse),nn.Sigmoid(),)defforward(self,x):b,c,_,_x.size()yself.avg_pool(x).view(b,c)y1self.fc1(y)y2self.fc2(y)y3self.fc3(y)y4self.fc4(y)y_concatetorch.cat([y1,y2,y3,y4],dim1)y_ex_dimself.fc(y_concate).view(b,c,1,1)returnx*y_ex_dim.expand_as(x)注册在ultralytics/nn/tasks.py中进行如下操作步骤1:fromultralytics.nn.attention.SENetV2importSaELayer步骤2修改def parse_model(d, ch, verboseTrue):elifmin{SaELayer}:args[ch[f],*args]配置yolov8_SENetV2.yamlultralytics/ultralytics/cfg/models/v8/yolov8_SENetV2.yaml# Ultralytics YOLO , AGPL-3.0 license# YOLOv8 object detection model with P3-P5 outputs. For Usage examples see https://docs.ultralytics.com/tasks/detect# Parametersnc:2# number of classesscales:# model compound scaling constants, i.e. modelyolov8n.yaml will call yolov8.yaml with scale n# [depth, width, max_channels]n:[0.33,0.25,1024]# YOLOv8n summary: 225 layers, 3157200 parameters, 3157184 gradients, 8.9 GFLOPss:[0.33,0.50,1024]# YOLOv8s summary: 225 layers, 11166560 parameters, 11166544 gradients, 28.8 GFLOPsm:[0.67,0.75,768]# YOLOv8m summary: 295 layers, 25902640 parameters, 25902624 gradients, 79.3 GFLOPsl:[1.00,1.00,512]# YOLOv8l summary: 365 layers, 43691520 parameters, 43691504 gradients, 165.7 GFLOPsx:[1.00,1.25,512]# YOLOv8x summary: 365 layers, 68229648 parameters, 68229632 gradients, 258.5 GFLOP# YOLOv8.0n backbonebackbone:# [from, repeats, module, args]-[-1,1,Conv,[64,3,2]]# 0-P1/2-[-1,1,Conv,[128,3,2]]# 1-P2/4-[-1,3,C2f,[128,True]]-[-1,1,Conv,[256,3,2]]# 3-P3/8-[-1,6,C2f,[256,True]]-[-1,1,Conv,[512,3,2]]# 5-P4/16-[-1,6,C2f,[512,True]]-[-1,1,Conv,[1024,3,2]]# 7-P5/32-[-1,3,C2f,[1024,True]]-[-1,1,SPPF,[1024,5]]# 9# YOLOv8.0n headhead:-[-1,1,nn.Upsample,[None,2,nearest]]-[[-1,6],1,Concat,[1]]# cat backbone P4-[-1,3,C2f,[512]]# 12-[-1,1,nn.Upsample,[None,2,nearest]]-[[-1,4],1,Concat,[1]]# cat backbone P3-[-1,3,C2f,[256]]# 15 (P3/8-small)-[-1,1,SaELayer,[]]# 16-[-1,1,Conv,[256,3,2]]-[[-1,12],1,Concat,[1]]# cat head P4-[-1,3,C2f,[512]]# 19 (P4/16-medium)-[-1,1,SaELayer,[]]# 20-[-1,1,Conv,[512,3,2]]-[[-1,9],1,Concat,[1]]# cat head P5-[-1,3,C2f,[1024]]# 23 (P5/32-large)-[-1,1,SaELayer,[]]# 24-[[16,20,24],1,Detect,[nc]]# Detect(P3, P4, P5)实验脚本importosfromultralyticsimportYOLO# Define the configuration options directlyyamlultralytics/cfg/models/v8/yolov8_SENetV2.yaml# Initialize the YOLO model with the specified YAML filemodelYOLO(yaml)# Print model informationmodel.info()if__name____main__:# Train the model with the specified parametersresultsmodel.train(dataultralytics/datasets/original-license-plates.yaml,nameSENetV2,epochs10,workers8,batch1)结果