GoogleNet

导入需要的包

import sys

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

sys.path.append(".")
import d2lzh_pytorch as d2l

Inception 块

Inception 块的基础结构为：

Inception 块是GoogleNet的基本组成单位，分为四条通道，也就是对同一个输入图像做四种处理，四条通道通过对填充与步幅的合理规定，可以使得四条通道得到的图像尺寸相同，最后将四个通道的输出在通道维上连结。

注意这里说的四条通道指的是四种处理方式，而非图像尺寸中的通道概念。最后的通道维才指的是图像尺寸中的通道。

Inception 块的超参数是各个卷积/池化层的输出通道数，对于多层的通道，用tuple结构定义某通道内各层的输出通道数。

定义代码如下：

class inception_block(nn.Module):
    def __init__(self, in_c, c1, c2: tuple, c3: tuple, c4):
        super(inception_block, self).__init__()
        # 左 1 通道
        self.p_1 = nn.Conv2d(in_c, c1, kernel_size=1)
        # 左 2 通道
        self.p_21 = nn.Conv2d(in_c, c2[0], kernel_size=1)
        self.p_22 = nn.Conv2d(c2[0], c2[1], kernel_size=3, padding=1)
        # 左 3 通道
        self.p_31 = nn.Conv2d(in_c, c3[0], kernel_size=1)
        self.p_32 = nn.Conv2d(c3[0], c3[1], kernel_size=5, padding=2)
        # 左 4 通道
        self.p_41 = nn.MaxPool2d(kernel_size=3, padding=1)
        self.p_42 = nn.Conv2d(in_c, c4, kernel_size=1)

    def forward(self, x):
        c1 = F.relu(self.p_1(x))
        c2 = F.relu(self.p_22(F.relu(self.p_21(x))))
        c3 = F.relu(self.p_32(F.relu(self.p_31(x))))
        c4 = F.relu(self.p_42(self.p_41(x)))
        return torch.cat((c1, c2, c3, c4), dim=1)

GoogleNet 网络

网络结构见原文档

代码如下：

b1 = nn.Sequential(
    nn.Conv2d(1, 64, kernel_size=7, stride=2, padding=3),
    nn.ReLU(),
    nn.MaxPool2d(kernel_size=3, stride=2, padding=1),
)

b2 = nn.Sequential(
    nn.Conv2d(64, 64, kernel_size=1),
    nn.Conv2d(64, 192, kernel_size=3, padding=1),
    nn.MaxPool2d(kernel_size=3, stride=2, padding=1),
)

b3 = nn.Sequential(
    Inception(192, 64, (96, 128), (16, 32), 32),
    Inception(256, 128, (128, 192), (32, 96), 64),
    nn.MaxPool2d(kernel_size=3, stride=2, padding=1),
)
b4 = nn.Sequential(
    Inception(480, 192, (96, 208), (16, 48), 64),
    Inception(512, 160, (112, 224), (24, 64), 64),
    Inception(512, 128, (128, 256), (24, 64), 64),
    Inception(512, 112, (144, 288), (32, 64), 64),
    Inception(528, 256, (160, 320), (32, 128), 128),
    nn.MaxPool2d(kernel_size=3, stride=2, padding=1),
)
b5 = nn.Sequential(
    Inception(832, 256, (160, 320), (32, 128), 128),
    Inception(832, 384, (192, 384), (48, 128), 128),
    d2l.GlobalAvgPool2d(),
)

net = nn.Sequential(b1, b2, b3, b4, b5, d2l.FlattenLayer(), nn.Linear(1024, 10))