GooLeNet模型搭建

一、model

import torch
from torch import nn
from torchsummary import summaryclass Inception(nn.Module):def __init__(self, in_channels, c1, c2 , c3 , c4):super(Inception, self).__init__()self.ReLU = nn.ReLU()#路线1:1x1卷积self.p1_1 = nn.Conv2d(in_channels = in_channels, out_channels = c1, kernel_size = 1)#路线2:1x1卷积->3x3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3:1x1卷积->5x5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4:3x3最大池化->1x1卷积self.p4_1 = nn.MaxPool2d(kernel_size = 3, stride = 1, padding = 1)self.p4_2 = nn.Conv2d(in_channels = in_channels, out_channels = c4, kernel_size = 1)#前向传播def forward(self, x):#路线1p1 = self.ReLU(self.p1_1(x))#路线2p2 = self.ReLU(self.p2_2(self.ReLU(self.p2_1(x))))#路线3p3 = self.ReLU(self.p3_2(self.ReLU(self.p3_1(x))))#路线4p4 = self.ReLU(self.p4_2(self.p4_1(x)))#拼接out = torch.cat((p1, p2, p3, p4), dim = 1)return outclass 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))#Inception模块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,(144,288),(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)),#平展层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 xif __name__ == '__main__':device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')model = GoogLeNet(Inception).to(device)print(summary(model,(1,224,224)))

二、model_train

import copy
import time
from torchvision.datasets import FashionMNIST
import numpy as np
from torchvision import transforms
import torch.utils.data as Data
import matplotlib.pyplot as plt
from model import GoogLeNet,Inception
import torch
import pandas as pddef train_val_data_process():# 数据预处理train_data = FashionMNIST(root='./data',train=True,download=True,transform=transforms.Compose([transforms.Resize(size=224),transforms.ToTensor()]))# 数据集划分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, #数据打乱num_workers=2)val_dataloader = Data.DataLoader(dataset=val_data,batch_size=32,shuffle=True,num_workers=2)#返回数据集return train_dataloader,val_dataloader# 训练模型
def train_model_process(model,train_dataloader,val_dataloader,num_epochs):device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')#定义优化器，optimizer = torch.optim.Adam(model.parameters(),lr=0.001)#定义损失函数(交叉熵损失函数)criterion = torch.nn.CrossEntropyLoss()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best_acc = 0.0#训练集与验证集损失函数列表train_loss_all = []val_loss_all = []#训练集与验证集准确率列表train_acc_all = []val_acc_all = []since = time.time()for epoch in range(num_epochs):print('Epoch {}/{}'.format(epoch, num_epochs - 1))print('-' * 10)# 每个epoch都有训练集训练过程和验证集评估过程train_loss = 0.0train_corrects = 0val_loss = 0.0val_corrects = 0# 训练集与验证集样本数量train_num = 0val_num = 0for step,(b_x,b_y) in enumerate(train_dataloader):#将特征和标签数据放入GPU中b_x = b_x.to(device)b_y = b_y.to(device)#设置模型为训练模式model.train()#前向传播，输入为一个batch的特征数据，输出为预测的标签output = model(b_x)#查看每一行中最大值所在的位置，即预测的类别pre_lab = torch.argmax(output,dim=1)# 计算损失函数loss = criterion(output,b_y)#将梯度置零，因为PyTorch中梯度是累加的optimizer.zero_grad()#反向传播，计算梯度loss.backward()#根据网络反向传播的梯度更新网络参数，达到降低loss的目的optimizer.step()#计算训练集的损失train_loss = train_loss + loss.item() * b_x.size(0)train_corrects = train_corrects + torch.sum(pre_lab == b_y.data)train_num = train_num + b_x.size(0)for step,(b_x,b_y) in enumerate(val_dataloader):# 将特征和标签数据放入验证设备中b_x = b_x.to(device)b_y = b_y.to(device)# 设置模型为评估模式model.eval()# 前向传播，输入为一个batch的特征数据，输出为预测的标签output = model(b_x)#查看每一行中最大值所在的位置，即预测的类别pre_lab = torch.argmax(output,dim=1)# 计算损失函数loss = criterion(output,b_y)# 计算验证集的损失val_loss = val_loss + loss.item() * b_x.size(0)val_corrects = val_corrects + torch.sum(pre_lab == b_y.data)val_num = val_num + b_x.size(0)# 计算训练集与测试集每个epoch的损失和准确率train_loss_all.append(train_loss / train_num)train_acc_all.append(train_corrects.double() / train_num)val_loss_all.append(val_loss / val_num)val_acc_all.append(val_corrects.double() / val_num)# 打印训练集和验证集准确率print('Train Loss: {:.4f} Acc: {:.4f}'.format(train_loss_all[-1], train_acc_all[-1]))print('Val Loss: {:.4f} Acc: {:.4f}'.format(val_loss_all[-1], val_acc_all[-1]))if val_acc_all[-1] > best_acc: # 保存准确率最高的模型参数best_acc = val_acc_all[-1]#保存当前的最高准确度best_model_wts = copy.deepcopy(model.state_dict())#训练耗时time_use = time.time() - sinceprint("训练和验证耗费的时间{:.0f}m{:.0f}s".format(time_use / 60,time_use % 60))#选择最优的模型参数#加载最高准确率的模型参数# model.load_state_dict(best_model_wts)# torch.save(model.load_state_dict(best_model_wts), 'best_model.pth')torch.save(best_model_wts, 'best_model.pth')train_porcess = pd.DataFrame(data = {'epoch':range(num_epochs),'train_loss_all':train_loss_all,'train_acc_all':train_acc_all,'val_loss_all':val_loss_all,'val_acc_all':val_acc_all})return train_porcessdef matplot_acc_loss(train_process):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"],'bo-',label="val_loss")plt.legend()plt.xlabel("epoch")plt.ylabel("loss")plt.figure(figsize=(12, 4))plt.subplot(1, 2, 1)plt.plot(train_process["epoch"], train_process["train_acc_all"], 'ro-', label="train_loss")plt.plot(train_process["epoch"], train_process["val_acc_all"], 'bo-', label="val_loss")plt.legend()plt.xlabel("epoch")plt.ylabel("acc")plt.show()if __name__=='__main__':#模型实例化GoogLeNet = GoogLeNet(Inception)train_dataloader,val_dataloader = train_val_data_process()train_porcess = train_model_process(GoogLeNet, train_dataloader,val_dataloader,num_epochs=10)matplot_acc_loss(train_porcess)

三、model_test

import torch
import torch.utils.data as Data
from torchvision import transforms
from torchvision.datasets import FashionMNIST
from model import GoogLeNet,Inception
import matplotlib.pyplot as plt
import numpy as np
import os
import timedef test_data_process():# 数据预处理test_data = FashionMNIST(root='./data',train=False,download=True,transform=transforms.Compose([transforms.Resize(size=224), transforms.ToTensor()]))test_dataloader = Data.DataLoader(dataset=test_data,batch_size=1,shuffle=True,  # 数据打乱num_workers=2)return test_dataloader#测试过程
def test_model_process(model,test_dataloader):device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')#将模型放入设备中model = model.to(device)#初始化参数test_corrects = 0test_num = 0#只进行前向传播，不计算梯度，节省内存，提高速度with torch.no_grad():for test_data_x,test_data_y in test_dataloader:test_data_x = test_data_x.to(device)test_data_y = test_data_y.to(device)#设置模型为评估模式model.eval()#前向传播,输入为测试数据集，输出为每个样本的预测值test_output = model(test_data_x)#获取预测值中最大值对应的索引，即预测的类别pre_lab = torch.argmax(test_output, 1)  #预测标签test_corrects += torch.sum(pre_lab == test_data_y.data)  #预测正确的数量test_num = test_num + test_data_y.size(0)  #测试样本数量#计算测试集的准确率test_acc = test_corrects.double().item() / test_num#打印测试集的准确率print('Test Accuracy: {:.4f}'.format(test_acc))if __name__ == '__main__':model = GoogLeNet(Inception)#加载训练好的模型model.load_state_dict(torch.load('best_model.pth'))test_dataloader = test_data_process() #加载测试数据# test_model_process(model,test_dataloader) #测试模型#模型推理device = 'cuda' if torch.cuda.is_available() else 'cpu'model = model.to(device)#将模型设置为评估模式with torch.no_grad():for b_x,b_y in test_dataloader:b_x = b_x.to(device)b_y = b_y.to(device)#模型设为验证模型model.eval()output = model(b_x)#获取预测值中最大值对应的索引，即预测的类别pre_lab = torch.argmax(output, dim = 1)  #预测标签result = pre_lab.item()label = b_y.item()#打印预测结果print('预测结果：',result,"--------","真实值:",label)