Pytorch 小记 第八回：GoogleNet卷积神经网络模型代码

本次小记，提供了一份基于pytorch的GoogleNet卷积神经网络模型的代码。除此之外，对代码中不容易理解的部分进行了讲解。
本代码的平台是PyCharm 2024.1.3，python版本3.11 numpy版本是1.26.4，pytorch版本2.0.0+cu118，d2l的版本是1.0.3

import numpy as np
import torch
from torch import nn
from torchvision.datasets import CIFAR10
import time
from torch.utils.data import DataLoader
from d2l import torch as d2l
import matplotlib.pyplot as plt


# 定义 inception块
class inception(nn.Module):
    def __init__(self, channel_in, channel_out1, channel_out2, channel_out3, channel_out4):
        super(inception, self).__init__()
        channel_out2_1 = channel_out2[0]
        channel_out2_2 = channel_out2[1]
        channel_out3_1 = channel_out3[0]
        channel_out3_2 = channel_out3[1]

        # 第一条线路 1*1卷积
        self.path1 = nn.Sequential(
            nn.Conv2d(channel_in, channel_out1, kernel_size=1, stride=1),
            nn.BatchNorm2d(channel_out1, eps=1e-3),
            nn.ReLU(True)
        )

        # 第二条线路 1*1卷积-> 3*3卷积+填充1
        self.path2 = nn.Sequential(
            nn.Conv2d(channel_in, channel_out2_1, kernel_size=1, stride=1),
            nn.BatchNorm2d(channel_out2_1, eps=1e-3),
            nn.ReLU(True),
            nn.Conv2d(channel_out2_1, channel_out2_2, kernel_size=3, stride=1, padding=1),
            nn.BatchNorm2d(channel_out2_2, eps=1e-3),
            nn.ReLU(True)
        )
        # 第三条线路 1*1卷积-> 5*5卷积+填充2
        self.path3 = nn.Sequential(
            nn.Conv2d(channel_in, channel_out3_1, kernel_size=1, stride=1),
            nn.BatchNorm2d(channel_out3_1, eps=1e-3),
            nn.ReLU(True),
            nn.Conv2d(channel_out3_1, channel_out3_2, kernel_size=5, stride=1, padding=2),
            nn.BatchNorm2d(channel_out3_2, eps=1e-3),
            nn.ReLU(True)
        )
        # 第四条线路 3*3池化+填充1-> 3*3卷积
        self.path4 = nn.Sequential(
            nn.MaxPool2d(3, stride=1, padding=1),
            nn.Conv2d(channel_in, channel_out4, kernel_size=1, stride=1),
            nn.BatchNorm2d(channel_out4, eps=1e-3),
            nn.ReLU(True)
        )

    def forward(self, x):
        y1 = self.path1(x)
        y2 = self.path2(x)
        y3 = self.path3(x)
        y4 = self.path4(x)
        output = torch.cat((y1, y2, y3,