当前位置： 首页 > news >正文

做网站标准步骤网络运营商

news 2025/7/6 0:49:12
做网站标准步骤,网络运营商,网站优化怎么做 有什么技巧,wordpress页面怎么写简介 Xception的名称源自于"Extreme Inception"，它是在Inception架构的基础上进行了扩展和改进。Inception架构是Google团队提出的一种经典的卷积神经网络架构，用于解决深度卷积神经网络中的计算和参数增长问题。 与Inception不同&#xff0…

简介

Xception的名称源自于"Extreme Inception"，它是在Inception架构的基础上进行了扩展和改进。Inception架构是Google团队提出的一种经典的卷积神经网络架构，用于解决深度卷积神经网络中的计算和参数增长问题。

与Inception不同，Xception的主要创新在于使用了深度可分离卷积（Depthwise Separable Convolution）来替代传统的卷积操作。深度可分离卷积将卷积操作分解为两个步骤：深度卷积和逐点卷积。

深度卷积是一种在每个输入通道上分别应用卷积核的操作，它可以有效地减少计算量和参数数量。逐点卷积是一种使用1x1卷积核进行通道间的线性组合的操作，用于增加模型的表示能力。通过使用深度可分离卷积，Xception网络能够更加有效地学习特征表示，并在相同计算复杂度下获得更好的性能。

Xception 网络结构

一个标准的Inception模块(Inception V3)

简化后的Inception模块

简化后的Inception的等价结构

采用深度可分离卷积的思想，使 3×3 卷积的数量与 1×1卷积输出通道的数量相等

Xception模型，一共可以分为3个flow，分别是Entry flow、Middle flow、Exit flow。

在这里 Entry 与 Exit 都具有相同的部分，Middle 与这二者有所不同。

Xception模型的pytorch复现

（1）深度可分离卷积

class SeparableConv2d(nn.Module):def __init__(self, in_channels, out_channels, kernel_size=3, stride=1, padding=0, dilation=1, bias=False):super(SeparableConv2d, self).__init__()self.conv = nn.Conv2d(in_channels, in_channels, kernel_size, stride, padding,dilation, groups=in_channels, bias=bias)self.pointwise = nn.Conv2d(in_channels, out_channels, kernel_size=1, stride=1, padding=0,dilation=1, groups=1, bias=False)def forward(self, x):x = self.conv(x)x = self.pointwise(x)return x

（2）构建三个flow结构

class EntryFlow(nn.Module):def __init__(self):super(EntryFlow, self).__init__()self.headconv = nn.Sequential(nn.Conv2d(3, 32, 3, 2, bias=False),nn.BatchNorm2d(32),nn.ReLU(inplace=True),nn.Conv2d(32, 64, 3, bias=False),nn.BatchNorm2d(64),nn.ReLU(inplace=True),)self.residual_block1 = nn.Sequential(SeparableConv2d(64, 128, 3, padding=1),nn.BatchNorm2d(128),nn.ReLU(inplace=True),SeparableConv2d(128, 128, 3, padding=1),nn.BatchNorm2d(128),nn.MaxPool2d(3, stride=2, padding=1),)self.residual_block2 = nn.Sequential(nn.ReLU(inplace=True),SeparableConv2d(128, 256, 3, padding=1),nn.BatchNorm2d(256),nn.ReLU(inplace=True),SeparableConv2d(256, 256, 3, padding=1),nn.BatchNorm2d(256),nn.MaxPool2d(3, stride=2, padding=1))self.residual_block3 = nn.Sequential(nn.ReLU(inplace=True),SeparableConv2d(256, 728, 3, padding=1),nn.BatchNorm2d(728),nn.ReLU(inplace=True),SeparableConv2d(728, 728, 3, padding=1),nn.BatchNorm2d(728),nn.MaxPool2d(3, stride=2, padding=1))def shortcut(self, inp, oup):return nn.Sequential(nn.Conv2d(inp, oup, 1, 2, bias=False),nn.BatchNorm2d(oup))def forward(self, x):x = self.headconv(x)residual = self.residual_block1(x)shortcut_block1 = self.shortcut(64, 128)x = residual + shortcut_block1(x)residual = self.residual_block2(x)shortcut_block2 = self.shortcut(128, 256)x = residual + shortcut_block2(x)residual = self.residual_block3(x)shortcut_block3 = self.shortcut(256, 728)x = residual + shortcut_block3(x)return xclass MiddleFlow(nn.Module):def __init__(self):super(MiddleFlow, self).__init__()self.shortcut = nn.Sequential()self.conv1 = nn.Sequential(nn.ReLU(inplace=True),SeparableConv2d(728, 728, 3, padding=1),nn.BatchNorm2d(728),nn.ReLU(inplace=True),SeparableConv2d(728, 728, 3, padding=1),nn.BatchNorm2d(728),nn.ReLU(inplace=True),SeparableConv2d(728, 728, 3, padding=1),nn.BatchNorm2d(728))def forward(self, x):residual = self.conv1(x)input = self.shortcut(x)return input + residualclass ExitFlow(nn.Module):def __init__(self):super(ExitFlow, self).__init__()self.residual_with_exit = nn.Sequential(nn.ReLU(inplace=True),SeparableConv2d(728, 728, 3, padding=1),nn.BatchNorm2d(728),nn.ReLU(inplace=True),SeparableConv2d(728, 1024, 3, padding=1),nn.BatchNorm2d(1024),nn.MaxPool2d(3, stride=2, padding=1))self.endconv = nn.Sequential(SeparableConv2d(1024, 1536, 3, 1, 1),nn.BatchNorm2d(1536),nn.ReLU(inplace=True),SeparableConv2d(1536, 2048, 3, 1, 1),nn.BatchNorm2d(2048),nn.ReLU(inplace=True),nn.AdaptiveAvgPool2d((1, 1)),)def shortcut(self, inp, oup):return nn.Sequential(nn.Conv2d(inp, oup, 1, 2, bias=False),nn.BatchNorm2d(oup))def forward(self, x):residual = self.residual_with_exit(x)shortcut_block = self.shortcut(728, 1024)output = residual + shortcut_block(x)return self.endconv(output)

（3）构建网络（完整代码）

"""
Copyright (c) 2023, Auorui.
All rights reserved.Xception: Deep Learning with Depthwise Separable Convolutions<https://arxiv.org/pdf/1610.02357.pdf>
"""
import torch
import torch.nn as nnclass SeparableConv2d(nn.Module):def __init__(self, in_channels, out_channels, kernel_size=3, stride=1, padding=0, dilation=1, bias=False):super(SeparableConv2d, self).__init__()self.conv = nn.Conv2d(in_channels, in_channels, kernel_size, stride, padding,dilation, groups=in_channels, bias=bias)self.pointwise = nn.Conv2d(in_channels, out_channels, kernel_size=1, stride=1, padding=0,dilation=1, groups=1, bias=False)def forward(self, x):x = self.conv(x)x = self.pointwise(x)return xclass EntryFlow(nn.Module):def __init__(self):super(EntryFlow, self).__init__()self.headconv = nn.Sequential(nn.Conv2d(3, 32, 3, 2, bias=False),nn.BatchNorm2d(32),nn.ReLU(inplace=True),nn.Conv2d(32, 64, 3, bias=False),nn.BatchNorm2d(64),nn.ReLU(inplace=True),)self.residual_block1 = nn.Sequential(SeparableConv2d(64, 128, 3, padding=1),nn.BatchNorm2d(128),nn.ReLU(inplace=True),SeparableConv2d(128, 128, 3, padding=1),nn.BatchNorm2d(128),nn.MaxPool2d(3, stride=2, padding=1),)self.residual_block2 = nn.Sequential(nn.ReLU(inplace=True),SeparableConv2d(128, 256, 3, padding=1),nn.BatchNorm2d(256),nn.ReLU(inplace=True),SeparableConv2d(256, 256, 3, padding=1),nn.BatchNorm2d(256),nn.MaxPool2d(3, stride=2, padding=1))self.residual_block3 = nn.Sequential(nn.ReLU(inplace=True),SeparableConv2d(256, 728, 3, padding=1),nn.BatchNorm2d(728),nn.ReLU(inplace=True),SeparableConv2d(728, 728, 3, padding=1),nn.BatchNorm2d(728),nn.MaxPool2d(3, stride=2, padding=1))def shortcut(self, inp, oup):return nn.Sequential(nn.Conv2d(inp, oup, 1, 2, bias=False),nn.BatchNorm2d(oup))def forward(self, x):x = self.headconv(x)residual = self.residual_block1(x)shortcut_block1 = self.shortcut(64, 128)x = residual + shortcut_block1(x)residual = self.residual_block2(x)shortcut_block2 = self.shortcut(128, 256)x = residual + shortcut_block2(x)residual = self.residual_block3(x)shortcut_block3 = self.shortcut(256, 728)x = residual + shortcut_block3(x)return xclass MiddleFlow(nn.Module):def __init__(self):super(MiddleFlow, self).__init__()self.shortcut = nn.Sequential()self.conv1 = nn.Sequential(nn.ReLU(inplace=True),SeparableConv2d(728, 728, 3, padding=1),nn.BatchNorm2d(728),nn.ReLU(inplace=True),SeparableConv2d(728, 728, 3, padding=1),nn.BatchNorm2d(728),nn.ReLU(inplace=True),SeparableConv2d(728, 728, 3, padding=1),nn.BatchNorm2d(728))def forward(self, x):residual = self.conv1(x)input = self.shortcut(x)return input + residualclass ExitFlow(nn.Module):def __init__(self):super(ExitFlow, self).__init__()self.residual_with_exit = nn.Sequential(nn.ReLU(inplace=True),SeparableConv2d(728, 728, 3, padding=1),nn.BatchNorm2d(728),nn.ReLU(inplace=True),SeparableConv2d(728, 1024, 3, padding=1),nn.BatchNorm2d(1024),nn.MaxPool2d(3, stride=2, padding=1))self.endconv = nn.Sequential(SeparableConv2d(1024, 1536, 3, 1, 1),nn.BatchNorm2d(1536),nn.ReLU(inplace=True),SeparableConv2d(1536, 2048, 3, 1, 1),nn.BatchNorm2d(2048),nn.ReLU(inplace=True),nn.AdaptiveAvgPool2d((1, 1)),)def shortcut(self, inp, oup):return nn.Sequential(nn.Conv2d(inp, oup, 1, 2, bias=False),nn.BatchNorm2d(oup))def forward(self, x):residual = self.residual_with_exit(x)shortcut_block = self.shortcut(728, 1024)output = residual + shortcut_block(x)return self.endconv(output)class Xception(nn.Module):def __init__(self, num_classes=1000):super().__init__()self.num_classes = num_classesself.entry_flow = EntryFlow()self.middle_flow = MiddleFlow()self.exit_flow = ExitFlow()self.fc = nn.Linear(2048, num_classes)def forward(self, x):x = self.entry_flow(x)for i in range(8):x = self.middle_flow(x)x = self.exit_flow(x)x = x.view(x.size(0), -1)out = self.fc(x)return outif __name__=='__main__':import torchsummarydevice = 'cuda' if torch.cuda.is_available() else 'cpu'input = torch.ones(2, 3, 224, 224).to(device)net = Xception(num_classes=4)net = net.to(device)out = net(input)print(out)print(out.shape)torchsummary.summary(net, input_size=(3, 224, 224))# Xception Total params: 19,838,076

参考文章

【精读AI论文】Xception ------(Xception: Deep Learning with Depthwise Separable Convolutions）_xception论文-CSDN博客

[ 轻量级网络 ] 经典网络模型4——Xception 详解与复现-CSDN博客

神经网络学习小记录22——Xception模型的复现详解_xception timm-CSDN博客

【卷积神经网络系列】十七、Xception_xception模块-CSDN博客 

http://www.15wanjia.com/news/9514.html

相关文章：

  • 热转印 东莞网站建设seo的培训班
  • 合肥市建设网站优化大师免费版
  • 云南省建设厅招标办网站seo详细教程
  • 常德网站建设wynet123seo智能优化系统
  • win7做本地网站电工培训学校
  • 泰州网站建设公司seo国外推广软件
  • 破解织梦做的网站长沙优化科技有限公司正规吗
  • 公司禁用网站怎么做百度帐号个人中心
  • 做网站需要的语言国内新闻
  • wordpress调用文章内容标签湖南优化推广
  • 小说网站建设目的什么是seo推广
  • 网站 建设需国际军事最新头条新闻
  • 寻找网站建设推广杭州网站推广与优化
  • 做建筑材料的网站有哪些重庆seo论坛
  • 石家庄网站制作武汉打开一个网站
  • 供应链网站制作注册网站需要多少钱?
  • 上海做网站搜索一下马来西亚的网络营销专业好就业吗
  • ASP做旅游网站代码三只松鼠网络营销策略
  • 网站建设免备案免费空间南京seo网站管理
  • 温州网站建设温州网站制作一键优化大师下载
  • 什么是品牌vi设计江门seo外包公司
  • centos7系统做网站网站备案
  • 个人做二次元网站怎么赚钱如何做推广
  • 学校诗歌网站建设推广费用一般多少钱
  • 外贸网站论文seo技术网网
  • 苏州网站建设品牌韶山百度seo
  • 网站建设中跳转页面源码竞价是什么意思
  • 深圳网站建设公司pestl分析软件培训机构有哪些?哪个比较好
  • 房地产网站素材重庆官网seo分析
  • 中山做展示型网站seo网站推广的主要目的