标签：kernel nn self v2 channels PyTorch Inception size out

PyTorch实现的Inception-v2
PyTorch: https://github.com/shanglianlm0525/ClassicNetwork

PyTorch代码:

import torch
import torch.nn as nn
import torchvision

def ConvBNReLU(in_channels,out_channels,kernel_size,stride=1,padding=0):
    return nn.Sequential(
        nn.Conv2d(in_channels=in_channels, out_channels=out_channels, kernel_size=kernel_size, stride=stride,padding=padding),
        nn.BatchNorm2d(out_channels),
        nn.ReLU6(inplace=True),
    )

def ConvBNReLUFactorization(in_channels,out_channels,kernel_sizes,paddings):
    return nn.Sequential(
        nn.Conv2d(in_channels=in_channels, out_channels=out_channels, kernel_size=kernel_sizes, stride=1,padding=paddings),
        nn.BatchNorm2d(out_channels),
        nn.ReLU6(inplace=True),
        nn.Conv2d(in_channels=out_channels, out_channels=out_channels, kernel_size=kernel_sizes, stride=1, padding=paddings),
        nn.BatchNorm2d(out_channels),
        nn.ReLU6(inplace=True),
    )

class InceptionV2ModuleA(nn.Module):
    def __init__(self, in_channels,out_channels1,out_channels2reduce, out_channels2, out_channels3reduce, out_channels3, out_channels4):
        super(InceptionV2ModuleA, self).__init__()

        self.branch1 = ConvBNReLU(in_channels=in_channels,out_channels=out_channels1,kernel_size=1)

        self.branch2 = nn.Sequential(
            ConvBNReLU(in_channels=in_channels, out_channels=out_channels2reduce, kernel_size=1),
            ConvBNReLU(in_channels=out_channels2reduce, out_channels=out_channels2, kernel_size=3, padding=1),
        )

        self.branch3 = nn.Sequential(
            ConvBNReLU(in_channels=in_channels,out_channels=out_channels3reduce,kernel_size=1),
            ConvBNReLU(in_channels=out_channels3reduce, out_channels=out_channels3, kernel_size=3, padding=1),
            ConvBNReLU(in_channels=out_channels3, out_channels=out_channels3, kernel_size=3, padding=1),
        )

        self.branch4 = nn.Sequential(
            nn.MaxPool2d(kernel_size=3, stride=1, padding=1),
            ConvBNReLU(in_channels=in_channels, out_channels=out_channels4, kernel_size=1),
        )

    def forward(self, x):
        out1 = self.branch1(x)
        out2 = self.branch2(x)
        out3 = self.branch3(x)
        out4 = self.branch4(x)
        out = torch.cat([out1, out2, out3, out4], dim=1)
        return out

class InceptionV2ModuleB(nn.Module):
    def __init__(self, in_channels,out_channels1,out_channels2reduce, out_channels2, out_channels3reduce, out_channels3, out_channels4):
        super(InceptionV2ModuleB, self).__init__()

        self.branch1 = ConvBNReLU(in_channels=in_channels,out_channels=out_channels1,kernel_size=1)

        self.branch2 = nn.Sequential(
            ConvBNReLU(in_channels=in_channels, out_channels=out_channels2reduce, kernel_size=1),
            ConvBNReLUFactorization(in_channels=out_channels2reduce, out_channels=out_channels2reduce, kernel_sizes=[1,3],paddings=[0,1]),
            ConvBNReLUFactorization(in_channels=out_channels2reduce, out_channels=out_channels2, kernel_sizes=[3,1],paddings=[1, 0]),
        )

        self.branch3 = nn.Sequential(
            ConvBNReLU(in_channels=in_channels,out_channels=out_channels3reduce,kernel_size=1),
            ConvBNReLUFactorization(in_channels=out_channels3reduce, out_channels=out_channels3reduce,kernel_sizes=[3, 1], paddings=[1, 0]),
            ConvBNReLUFactorization(in_channels=out_channels3reduce, out_channels=out_channels3reduce,kernel_sizes=[1, 3], paddings=[0, 1]),
            ConvBNReLUFactorization(in_channels=out_channels3reduce, out_channels=out_channels3reduce,kernel_sizes=[3, 1], paddings=[1, 0]),
            ConvBNReLUFactorization(in_channels=out_channels3reduce, out_channels=out_channels3,kernel_sizes=[1, 3], paddings=[0, 1]),
        )

        self.branch4 = nn.Sequential(
            nn.MaxPool2d(kernel_size=3, stride=1, padding=1),
            ConvBNReLU(in_channels=in_channels, out_channels=out_channels4, kernel_size=1),
        )

    def forward(self, x):
        out1 = self.branch1(x)
        out2 = self.branch2(x)
        out3 = self.branch3(x)
        out4 = self.branch4(x)
        out = torch.cat([out1, out2, out3, out4], dim=1)
        return out

class InceptionV2ModuleC(nn.Module):
    def __init__(self, in_channels,out_channels1,out_channels2reduce, out_channels2, out_channels3reduce, out_channels3, out_channels4):
        super(InceptionV2ModuleC, self).__init__()

        self.branch1 = ConvBNReLU(in_channels=in_channels,out_channels=out_channels1,kernel_size=1)

        self.branch2_conv1 = ConvBNReLU(in_channels=in_channels, out_channels=out_channels2reduce, kernel_size=1)
        self.branch2_conv2a = ConvBNReLUFactorization(in_channels=out_channels2reduce, out_channels=out_channels2, kernel_sizes=[1,3],paddings=[0,1])
        self.branch2_conv2b = ConvBNReLUFactorization(in_channels=out_channels2reduce, out_channels=out_channels2, kernel_sizes=[3,1],paddings=[1,0])

        self.branch3_conv1 = ConvBNReLU(in_channels=in_channels,out_channels=out_channels3reduce,kernel_size=1)
        self.branch3_conv2 = ConvBNReLU(in_channels=out_channels3reduce, out_channels=out_channels3, kernel_size=3,stride=1,padding=1)
        self.branch3_conv3a = ConvBNReLUFactorization(in_channels=out_channels3, out_channels=out_channels3, kernel_sizes=[3, 1],paddings=[1, 0])
        self.branch3_conv3b = ConvBNReLUFactorization(in_channels=out_channels3, out_channels=out_channels3, kernel_sizes=[1, 3],paddings=[0, 1])

        self.branch4 = nn.Sequential(
            nn.MaxPool2d(kernel_size=3, stride=1, padding=1),
            ConvBNReLU(in_channels=in_channels, out_channels=out_channels4, kernel_size=1),
        )

    def forward(self, x):
        out1 = self.branch1(x)
        x2 = self.branch2_conv1(x)
        out2 = torch.cat([self.branch2_conv2a(x2), self.branch2_conv2b(x2)],dim=1)
        x3 = self.branch3_conv2(self.branch3_conv1(x))
        out3 = torch.cat([self.branch3_conv3a(x3), self.branch3_conv3b(x3)], dim=1)
        out4 = self.branch4(x)
        out = torch.cat([out1, out2, out3, out4], dim=1)
        return out

class InceptionV3ModuleD(nn.Module):
    def __init__(self, in_channels,out_channels1reduce,out_channels1,out_channels2reduce, out_channels2):
        super(InceptionV3ModuleD, self).__init__()

        self.branch1 = nn.Sequential(
            ConvBNReLU(in_channels=in_channels, out_channels=out_channels1reduce, kernel_size=1),
            ConvBNReLU(in_channels=out_channels1reduce, out_channels=out_channels1, kernel_size=3,stride=2,padding=1)
        )

        self.branch2 = nn.Sequential(
            ConvBNReLU(in_channels=in_channels, out_channels=out_channels2reduce, kernel_size=1),
            ConvBNReLU(in_channels=out_channels2reduce, out_channels=out_channels2, kernel_size=3, stride=1, padding=1),
            ConvBNReLU(in_channels=out_channels2, out_channels=out_channels2, kernel_size=3, stride=2,padding=1),
        )

        self.branch3 = nn.MaxPool2d(kernel_size=3,stride=2,padding=1)

    def forward(self, x):
        out1 = self.branch1(x)
        out2 = self.branch2(x)
        out3 = self.branch3(x)
        out = torch.cat([out1, out2, out3], dim=1)
        return out

class InceptionAux(nn.Module):
    def __init__(self, in_channels,out_channels):
        super(InceptionAux, self).__init__()

        self.auxiliary_avgpool = nn.AvgPool2d(kernel_size=5, stride=3)
        self.auxiliary_conv1 = ConvBNReLU(in_channels=in_channels, out_channels=128, kernel_size=1)
        self.auxiliary_conv2 = nn.Conv2d(in_channels=128, out_channels=768, kernel_size=5,stride=1)
        self.auxiliary_dropout = nn.Dropout(p=0.7)
        self.auxiliary_linear1 = nn.Linear(in_features=768, out_features=out_channels)

    def forward(self, x):
        x = self.auxiliary_conv1(self.auxiliary_avgpool(x))
        x = self.auxiliary_conv2(x)
        x = x.view(x.size(0), -1)
        out = self.auxiliary_linear1(self.auxiliary_dropout(x))
        return out

class InceptionV2(nn.Module):
    def __init__(self, num_classes=1000, stage='train'):
        super(InceptionV2, self).__init__()
        self.stage = stage

        self.block1 = nn.Sequential(
            ConvBNReLU(in_channels=3, out_channels=64, kernel_size=7,stride=2,padding=3),
            nn.MaxPool2d(kernel_size=3,stride=2,padding=1),
        )

        self.block2 = nn.Sequential(
            ConvBNReLU(in_channels=64, out_channels=192, kernel_size=3, stride=1, padding=1),
            nn.MaxPool2d(kernel_size=3, stride=2,padding=1),
        )

        self.block3 = nn.Sequential(
            InceptionV2ModuleA(in_channels=192,out_channels1=64,out_channels2reduce=64, out_channels2=64, out_channels3reduce=64, out_channels3=96, out_channels4=32),
            InceptionV2ModuleA(in_channels=256, out_channels1=64, out_channels2reduce=64, out_channels2=96,out_channels3reduce=64, out_channels3=96, out_channels4=64),
            InceptionV3ModuleD(in_channels=320, out_channels1reduce=128, out_channels1=160, out_channels2reduce=64,out_channels2=96),
        )

        self.block4 = nn.Sequential(
            InceptionV2ModuleB(in_channels=576, out_channels1=224, out_channels2reduce=64, out_channels2=96,out_channels3reduce=96, out_channels3=128, out_channels4=128),
            InceptionV2ModuleB(in_channels=576, out_channels1=192, out_channels2reduce=96, out_channels2=128,out_channels3reduce=96, out_channels3=128, out_channels4=128),
            InceptionV2ModuleB(in_channels=576, out_channels1=160, out_channels2reduce=128, out_channels2=160,out_channels3reduce=128, out_channels3=128, out_channels4=128),
            InceptionV2ModuleB(in_channels=576, out_channels1=96, out_channels2reduce=128, out_channels2=192,out_channels3reduce=160, out_channels3=160, out_channels4=128),
            InceptionV3ModuleD(in_channels=576, out_channels1reduce=128, out_channels1=192, out_channels2reduce=192,out_channels2=256),
        )

        self.block5 = nn.Sequential(
            InceptionV2ModuleC(in_channels=1024, out_channels1=352, out_channels2reduce=192, out_channels2=160,out_channels3reduce=160, out_channels3=112, out_channels4=128),
            InceptionV2ModuleC(in_channels=1024, out_channels1=352, out_channels2reduce=192, out_channels2=160,
                               out_channels3reduce=192, out_channels3=112, out_channels4=128)
        )

        self.max_pool = nn.MaxPool2d(kernel_size=7, stride=1)
        self.dropout = nn.Dropout(p=0.5)
        self.linear = nn.Linear(1024, num_classes)

    def forward(self, x):
        x = self.block1(x)
        x = self.block2(x)
        x = self.block3(x)
        x = self.block4(x)
        x = self.block5(x)
        x = self.max_pool(x)
        x = self.dropout(x)
        x = x.view(x.size(0), -1)
        out = self.linear(x)
        return out

if __name__=='__main__':
    model = InceptionV2()
    print(model)

    input = torch.randn(1, 3, 224, 224)
    out = model(input)
    print(out.shape)

标签：kernel,nn,self,v2,channels,PyTorch,Inception,size,out
来源： https://blog.csdn.net/shanglianlm/article/details/99132682

本站声明： 1. iCode9 技术分享网（下文简称本站）提供的所有内容，仅供技术学习、探讨和分享；
2. 关于本站的所有留言、评论、转载及引用，纯属内容发起人的个人观点，与本站观点和立场无关；
3. 关于本站的所有言论和文字，纯属内容发起人的个人观点，与本站观点和立场无关；
4. 本站文章均是网友提供，不完全保证技术分享内容的完整性、准确性、时效性、风险性和版权归属；如您发现该文章侵犯了您的权益，可联系我们第一时间进行删除；
5. 本站为非盈利性的个人网站，所有内容不会用来进行牟利，也不会利用任何形式的广告来间接获益，纯粹是为了广大技术爱好者提供技术内容和技术思想的分享性交流网站。