pytorch-googleNet

googleNet網絡結構 

 

 輸入網絡: 由4個分支網絡構成

第一分支: 由1x1的卷積構成

第二分支: 由1x1的卷積，3x3的卷積構成

第三分支: 由1x1的卷積, 5x5的卷積構成

第四分支: 由3x3的最大值池化, 1x1的卷積構成 

import torch
from torch import nn
from torch.nn import functional as F

class BasicConv2d(nn.Module):
    def __init__(self, in_channels, out_channels, **kwargs):
        super(BasicConv2d, self).__init__()
        self.conv = nn.Conv2d(in_channels, out_channels, bias=False, **kwargs) # 構造卷積層
        self.bn = nn.BatchNorm2d(out_channels, eps=0.001) # 構造標準化

    def forward(self, x):
        x = self.conv(x) # 進行卷積操做
        x = self.bn(x) # 進行標準化操做
        x = F.relu(x) # 進行激活層操做

        return x

class Inception(nn.Module):
    def __init__(self, in_channels, pool_features):
        super(Inception, self).__init__()
        self.branch1x1 = BasicConv2d(in_channels, 64, kernel_size=1) # 1x1的卷積操做

        self.branch5x5_1 = BasicConv2d(in_channels, 48, kernel_size=1) # 進行卷積操做
        self.branch5x5_2 = BasicConv2d(48, 64, kernel_size=5, padding=2) 

        self.branch3x3db1_1 = BasicConv2d(in_channels, 64, kernel_size=1)
        self.branch3x3db1_2 = BasicConv2d(64, 96, kernel_size=3)
        self.branch3x3db1_3 = BasicConv2d(96, 96, kernel_size=3)

        self.branch_pool = BasicConv2d(in_channels, pool_features, kernel_size=1)

    def forward(self, x):
        branch1x1 = self.branch1x1(x)

        branch5x5 = self.branch5x5_1(x)
        branch5x5 = self.branch5x5_2(branch5x5)

        branch3x3db1_1 = self.branch3x3db1_1(x)
        branch3x3db1_2 = self.branch3x3db1_2(branch3x3db1_1)
        branch3x3db1_3 = self.branch3x3db1_3(branch3x3db1_2)

        branch_pool = F.max_pool2d(x, kernel_size=3, stride=1, padding=1)
        branch_pool = self.branch_pool(branch_pool)
        # 進行卷積的疊加操做
        outputs = [branch1x1, branch5x5, branch3x3db1_3, branch_pool]
        outputs = torch.cat(outputs, dim=1)

        return outputs