1.GoolgeNet代码分析
1.1Inception模块
class Inception(nn.Module): def __init__(self, in_channels, c1, c2, c3, c4): super(Inception, self).__init__() self.ReLU = nn.ReLU() # 路线1，单1×1卷积层 self.p1_1 = nn.Conv2d(in_channels=in_channels, out_channels=c1, kernel_size=1) # 路线2，1×1卷积层, 3×3的卷积 self.p2_1 = nn.Conv2d(in_channels=in_channels, out_channels=c2[0], kernel_size=1) self.p2_2 = nn.Conv2d(in_channels=c2[0], out_channels=c2[1], kernel_size=3, padding=1) # 路线3，1×1卷积层, 5×5的卷积 self.p3_1 = nn.Conv2d(in_channels=in_channels, out_channels=c3[0], kernel_size=1) self.p3_2 = nn.Conv2d(in_channels=c3[0], out_channels=c3[1], kernel_size=5, padding=2) # 路线4，3×3的最大池化, 1×1的卷积 self.p4_1 = nn.MaxPool2d(kernel_size=3, padding=1, stride=1) self.p4_2 = nn.Conv2d(in_channels=in_channels, out_channels=c4, kernel_size=1) def forward(self, x): p1 = self.ReLU(self.p1_1(x)) p2 = self.ReLU(self.p2_2(self.ReLU(self.p2_1(x)))) p3 = self.ReLU(self.p3_2(self.ReLU(self.p3_1(x)))) p4 = self.ReLU(self.p4_2(self.p4_1(x))) return torch.cat((p1, p2, p3, p4), dim=1)这里Inception模块包含四条并行路径：
路径1：1×1卷积 (降维)路径2：1×1卷积 → 3×3卷积 (提取中等特征)
路径3：1×1卷积 → 5×5卷积 (提取大范围特征)
路径4：3×3最大池化 → 1×1卷积 (保留空间信息)
1.2GoogLeNet主体结构
class GoogLeNet(nn.Module): def __init__(self, Inception): super(GoogLeNet, self).__init__() self.b1 = nn.Sequential( nn.Conv2d(in_channels=1, out_channels=64, kernel_size=7, stride=2, padding=3), nn.ReLU(), nn.MaxPool2d(kernel_size=3, stride=2, padding=1)) self.b2 = nn.Sequential( nn.Conv2d(in_channels=64, out_channels=64, kernel_size=1), nn.ReLU(), nn.Conv2d(in_channels=64, out_channels=192, kernel_size=3, padding=1), nn.ReLU(), nn.MaxPool2d(kernel_size=3, stride=2, padding=1)) self.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)) self.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, (128, 288), (32, 64), 64), Inception(528, 256, (160, 320), (32, 128), 128), nn.MaxPool2d(kernel_size=3, stride=2, padding=1)) self.b5 = nn.Sequential( Inception(832, 256, (160, 320), (32, 128), 128), Inception(832, 384, (192, 384), (48, 128), 128), nn.AdaptiveAvgPool2d((1, 1)), nn.Flatten(), nn.Linear(1024, 10)) for m in self.modules(): if isinstance(m, nn.Conv2d): nn.init.kaiming_normal_(m.weight, mode="fan_out", nonlinearity='relu') if m.bias is not None: nn.init.constant_(m.bias, 0) elif isinstance(m, nn.Linear): nn.init.normal_(m.weight, 0, 0.01) if m.bias is not None: nn.init.constant_(m.bias, 0) def forward(self, x): x = self.b1(x) x = self.b2(x) x = self.b3(x) x = self.b4(x) x = self.b5(x) return x其中这段是网络的参数初始化代码：
for m in self.modules(): if isinstance(m, nn.Conv2d): nn.init.kaiming_normal_(m.weight, mode="fan_out", nonlinearity='relu') if m.bias is not None: nn.init.constant_(m.bias, 0) elif isinstance(m, nn.Linear): nn.init.normal_(m.weight, 0, 0.01) if m.bias is not None: nn.init.constant_(m.bias, 0)这段代码就是遍历所有模块，判断是卷积层还是全连接层，然后进行相应的操作。
代码保证了卷积层使用适合ReLU的Kaiming初始化，所有偏置置于0，全连接层使用小标准差的正态分布初始化。
2.代码
#model import torch from torch import nn from torchsummary import summary class Inception(nn.Module): def __init__(self, in_channels, c1, c2, c3, c4): super(Inception, self).__init__() self.ReLU = nn.ReLU() # 路线1，单1×1卷积层 self.p1_1 = nn.Conv2d(in_channels=in_channels, out_channels=c1, kernel_size=1) # 路线2，1×1卷积层, 3×3的卷积 self.p2_1 = nn.Conv2d(in_channels=in_channels, out_channels=c2[0], kernel_size=1) self.p2_2 = nn.Conv2d(in_channels=c2[0], out_channels=c2[1], kernel_size=3, padding=1) # 路线3，1×1卷积层, 5×5的卷积 self.p3_1 = nn.Conv2d(in_channels=in_channels, out_channels=c3[0], kernel_size=1) self.p3_2 = nn.Conv2d(in_channels=c3[0], out_channels=c3[1], kernel_size=5, padding=2) # 路线4，3×3的最大池化, 1×1的卷积 self.p4_1 = nn.MaxPool2d(kernel_size=3, padding=1, stride=1) self.p4_2 = nn.Conv2d(in_channels=in_channels, out_channels=c4, kernel_size=1) def forward(self, x): p1 = self.ReLU(self.p1_1(x)) p2 = self.ReLU(self.p2_2(self.ReLU(self.p2_1(x)))) p3 = self.ReLU(self.p3_2(self.ReLU(self.p3_1(x)))) p4 = self.ReLU(self.p4_2(self.p4_1(x))) return torch.cat((p1, p2, p3, p4), dim=1) class GoogLeNet(nn.Module): def __init__(self, Inception): super(GoogLeNet, self).__init__() self.b1 = nn.Sequential( nn.Conv2d(in_channels=1, out_channels=64, kernel_size=7, stride=2, padding=3), nn.ReLU(), nn.MaxPool2d(kernel_size=3, stride=2, padding=1)) self.b2 = nn.Sequential( nn.Conv2d(in_channels=64, out_channels=64, kernel_size=1), nn.ReLU(), nn.Conv2d(in_channels=64, out_channels=192, kernel_size=3, padding=1), nn.ReLU(), nn.MaxPool2d(kernel_size=3, stride=2, padding=1)) self.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)) self.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, (128, 288), (32, 64), 64), Inception(528, 256, (160, 320), (32, 128), 128), nn.MaxPool2d(kernel_size=3, stride=2, padding=1)) self.b5 = nn.Sequential( Inception(832, 256, (160, 320), (32, 128), 128), Inception(832, 384, (192, 384), (48, 128), 128), nn.AdaptiveAvgPool2d((1, 1)), nn.Flatten(), nn.Linear(1024, 10)) for m in self.modules(): if isinstance(m, nn.Conv2d): nn.init.kaiming_normal_(m.weight, mode="fan_out", nonlinearity='relu') if m.bias is not None: nn.init.constant_(m.bias, 0) elif isinstance(m, nn.Linear): nn.init.normal_(m.weight, 0, 0.01) if m.bias is not None: nn.init.constant_(m.bias, 0) def forward(self, x): x = self.b1(x) x = self.b2(x) x = self.b3(x) x = self.b4(x) x = self.b5(x) return x if __name__ == "__main__": device = torch.device("cuda" if torch.cuda.is_available() else "cpu") model = GoogLeNet(Inception).to(device) print(summary(model, (1, 224, 224)))#model-train import copy import time import torch from torchvision.datasets import FashionMNIST from torchvision import transforms import torch.utils.data as Data import numpy as np import matplotlib.pyplot as plt from model import GoogLeNet, Inception import torch.nn as nn import pandas as pd def train_val_data_process(): train_data = FashionMNIST(root='./data', train=True, transform=transforms.Compose([transforms.Resize(size=224), transforms.ToTensor()]), download=True) train_data, val_data = Data.random_split(train_data, [round(0.8*len(train_data)), round(0.2*len(train_data))]) train_dataloader = Data.DataLoader(dataset=train_data, batch_size=32, shuffle=True, num_workers=2) val_dataloader = Data.DataLoader(dataset=val_data, batch_size=32, shuffle=True, num_workers=2) return train_dataloader, val_dataloader def train_model_process(model, train_dataloader, val_dataloader, num_epochs): # 设定训练所用到的设备，有GPU用GPU没有GPU用CPU device = torch.device("cuda" if torch.cuda.is_available() else "cpu") # 使用Adam优化器，学习率为0.001 optimizer = torch.optim.Adam(model.parameters(), lr=0.001) # 损失函数为交叉熵函数 criterion = nn.CrossEntropyLoss() # 将模型放入到训练设备中 model = model.to(device) # 复制当前模型的参数 best_model_wts = copy.deepcopy(model.state_dict()) # 初始化参数 # 最高准确度 best_acc = 0.0 # 训练集损失列表 train_loss_all = [] # 验证集损失列表 val_loss_all = [] # 训练集准确度列表 train_acc_all = [] # 验证集准确度列表 val_acc_all = [] # 当前时间 since = time.time() for epoch in range(num_epochs): print("Epoch {}/{}".format(epoch, num_epochs-1)) print("-"*10) # 初始化参数 # 训练集损失函数 train_loss = 0.0 # 训练集准确度 train_corrects = 0 # 验证集损失函数 val_loss = 0.0 # 验证集准确度 val_corrects = 0 # 训练集样本数量 train_num = 0 # 验证集样本数量 val_num = 0 # 对每一个mini-batch训练和计算 for step, (b_x, b_y) in enumerate(train_dataloader): # 将特征放入到训练设备中 b_x = b_x.to(device) # 将标签放入到训练设备中 b_y = b_y.to(device) # 设置模型为训练模式 model.train() # 前向传播过程，输入为一个batch，输出为一个batch中对应的预测 output = model(b_x) # 查找每一行中最大值对应的行标 pre_lab = torch.argmax(output, dim=1) # 计算每一个batch的损失函数 loss = criterion(output, b_y) # 将梯度初始化为0 optimizer.zero_grad() # 反向传播计算 loss.backward() # 根据网络反向传播的梯度信息来更新网络的参数，以起到降低loss函数计算值的作用 optimizer.step() # 对损失函数进行累加 train_loss += loss.item() * b_x.size(0) # 如果预测正确，则准确度train_corrects加1 train_corrects += torch.sum(pre_lab == b_y.data) # 当前用于训练的样本数量 train_num += b_x.size(0) for step, (b_x, b_y) in enumerate(val_dataloader): # 将特征放入到验证设备中 b_x = b_x.to(device) # 将标签放入到验证设备中 b_y = b_y.to(device) # 设置模型为评估模式 model.eval() # 前向传播过程，输入为一个batch，输出为一个batch中对应的预测 output = model(b_x) # 查找每一行中最大值对应的行标 pre_lab = torch.argmax(output, dim=1) # 计算每一个batch的损失函数 loss = criterion(output, b_y) # 对损失函数进行累加 val_loss += loss.item() * b_x.size(0) # 如果预测正确，则准确度train_corrects加1 val_corrects += torch.sum(pre_lab == b_y.data) # 当前用于验证的样本数量 val_num += b_x.size(0) # 计算并保存每一次迭代的loss值和准确率 # 计算并保存训练集的loss值 train_loss_all.append(train_loss / train_num) # 计算并保存训练集的准确率 train_acc_all.append(train_corrects.double().item() / train_num) # 计算并保存验证集的loss值 val_loss_all.append(val_loss / val_num) # 计算并保存验证集的准确率 val_acc_all.append(val_corrects.double().item() / val_num) print("{} train loss:{:.4f} train acc: {:.4f}".format(epoch, train_loss_all[-1], train_acc_all[-1])) print("{} val loss:{:.4f} val acc: {:.4f}".format(epoch, val_loss_all[-1], val_acc_all[-1])) if val_acc_all[-1] > best_acc: # 保存当前最高准确度 best_acc = val_acc_all[-1] # 保存当前最高准确度的模型参数 best_model_wts = copy.deepcopy(model.state_dict()) # 计算训练和验证的耗时 time_use = time.time() - since print("训练和验证耗费的时间{:.0f}m{:.0f}s".format(time_use//60, time_use%60)) # 选择最优参数，保存最优参数的模型 model.load_state_dict(best_model_wts) # torch.save(model.load_state_dict(best_model_wts), "C:/Users/86159/Desktop/LeNet/best_model.pth") torch.save(best_model_wts, "C:/Users/12072/Desktop/TEST/GoogLeNet/best_model.pth") train_process = pd.DataFrame(data={"epoch":range(num_epochs), "train_loss_all":train_loss_all, "val_loss_all":val_loss_all, "train_acc_all":train_acc_all, "val_acc_all":val_acc_all,}) return train_process def matplot_acc_loss(train_process): # 显示每一次迭代后的训练集和验证集的损失函数和准确率 plt.figure(figsize=(12, 4)) plt.subplot(1, 2, 1) plt.plot(train_process['epoch'], train_process.train_loss_all, "ro-", label="Train loss") plt.plot(train_process['epoch'], train_process.val_loss_all, "bs-", label="Val loss") plt.legend() plt.xlabel("epoch") plt.ylabel("Loss") plt.subplot(1, 2, 2) plt.plot(train_process['epoch'], train_process.train_acc_all, "ro-", label="Train acc") plt.plot(train_process['epoch'], train_process.val_acc_all, "bs-", label="Val acc") plt.xlabel("epoch") plt.ylabel("acc") plt.legend() plt.show() if __name__ == '__main__': # 加载需要的模型 GoogLeNet = GoogLeNet(Inception) # 加载数据集 train_data, val_data = train_val_data_process() # 利用现有的模型进行模型的训练 train_process = train_model_process(GoogLeNet, train_data, val_data, num_epochs=20) matplot_acc_loss(train_process)#model-test import torch import torch.utils.data as Data from torchvision import transforms from torchvision.datasets import FashionMNIST from model import GoogLeNet, Inception def test_data_process(): test_data = FashionMNIST(root='./data', train=False, transform=transforms.Compose([transforms.Resize(size=224), transforms.ToTensor()]), download=True) test_dataloader = Data.DataLoader(dataset=test_data, batch_size=1, shuffle=True, num_workers=0) return test_dataloader def test_model_process(model, test_dataloader): # 设定测试所用到的设备，有GPU用GPU没有GPU用CPU device = "cuda" if torch.cuda.is_available() else 'cpu' # 讲模型放入到训练设备中 model = model.to(device) # 初始化参数 test_corrects = 0.0 test_num = 0 # 只进行前向传播计算，不计算梯度，从而节省内存，加快运行速度 with torch.no_grad(): for test_data_x, test_data_y in test_dataloader: # 将特征放入到测试设备中 test_data_x = test_data_x.to(device) # 将标签放入到测试设备中 test_data_y = test_data_y.to(device) # 设置模型为评估模式 model.eval() # 前向传播过程，输入为测试数据集，输出为对每个样本的预测值 output= model(test_data_x) # 查找每一行中最大值对应的行标 pre_lab = torch.argmax(output, dim=1) # 如果预测正确，则准确度test_corrects加1 test_corrects += torch.sum(pre_lab == test_data_y.data) # 将所有的测试样本进行累加 test_num += test_data_x.size(0) # 计算测试准确率 test_acc = test_corrects.double().item() / test_num print("测试的准确率为：", test_acc) if __name__ == "__main__": # 加载模型 model = GoogLeNet(Inception) model.load_state_dict(torch.load('best_model.pth')) # # 利用现有的模型进行模型的测试 test_dataloader = test_data_process() test_model_process(model, test_dataloader) # 设定测试所用到的设备，有GPU用GPU没有GPU用CPU device = "cuda" if torch.cuda.is_available() else 'cpu' model = model.to(device) classes = ['T-shirt/top', 'Trouser', 'Pullover', 'Dress', 'Coat', 'Sandal', 'Shirt', 'Sneaker', 'Bag', 'Ankle boot'] with torch.no_grad(): for b_x, b_y in test_dataloader: b_x = b_x.to(device) b_y = b_y.to(device) # 设置模型为验证模型 model.eval() output = model(b_x) pre_lab = torch.argmax(output, dim=1) result = pre_lab.item() label = b_y.item() print("预测值：", classes[result], "------", "真实值：", classes[label])3.结果展示
model-train运行结果：
model-test运行结果：