Pytorch学习(十二)--Inception Module

传送门：Inception_Module
本讲视频是对高级CNN进行的一个概览，主要实现了Inception，Resnet，本文专注于Inception模块以及GoogLeNet网络的实现。

Inception Moudel代码说明:

1、采用1x1卷积核将不同通道的信息融合。使用1x1卷积核虽然参数量增加了，但是能够显著的降低计算量。
2、Inception模块分为不同的分支是为了提取到更多的特征，同时也可以加速训练过程。
3、Inception Moudel由4个分支组成，最终4个分支在dim=1(channels)上进行组合。最终输出的通道总数为24+16+24+24 = 88

Inception模块是现在代码如下，各分支对应图中标记：

class Inception(torch.nn.Module):
    def __init__(self,in_channels):
        super(Inception,self).__init__()
        # 定义分支1
        self.branch1x1 = nn.Conv2D(in_channels,16,kernel_size = 1)
        # 定义分支2
        self.branch5x5_1 = nn.Conv2D(in_channels,16,kernel_size = 1)
        self.branch5x5_2 = nn.Conv2D(16,24,kernel_size = 5,padding =1)
        # 定义分支3
        self.branch3x3_1 = nn.Conv2D(in_channels,16,kernel_size=1)
        self.branch3x3_2 = nn.Conv2D(16,24,kernel_size=  3,padding =1)
        self.branch3x3_3 = nn.Conv2D(24,24,kernel_size=  3,padding =1)
        # 定义分支4
        self.branch_pool = nn.Conv2D(in_channels,24,kernel_size = 1)

    def forward(self,x):
        # 计算分支1
        branch1x1 = self.branch1x1(x)
        # 计算分支2
        branch5x5 = self.branch5x5_1(x)
        branch5x5=self.branch5x5_2(branch5x5)
        # 计算分支4
        branch3x3 = self.branch3x3_1(x)
        branch3x3 = self.branch3x3_2(branch3x3)
        branch3x3 = self.branch3x3_3(branch3x3)
        # 计算分支4
        branch_pool = F.avg_pool2d(x, kernel_size=3, stride=1, padding=1)
        branch_pool = self.branch_pool(branch_pool)

        #拼接四个不同分支得到的通道，作为输出
        outputs = [branch1x1, branch5x5, branch3x3, branch_pool]
        return torch.cat(outputs, dim=1)  # b,c,w,h  c对应的是dim=1

使用Inception模块编写网络，代码如下：

class Net(torch.nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        #两个卷积层
        self.conv1 = torch.nn.Conv2d(1, 10, kernel_size=5)
        self.conv2 = torch.nn.Conv2d(88, 20, kernel_size=5)  # 88 = 24x3 + 16

        #两个Inception模块
        self.incep1 = Inception(in_channels=10)  # 输入为10，与conv1 中的10对应
        self.incep2 = Inception(in_channels=20)  # 输入为20，与conv2 中的20对应

        self.mp = torch.nn.MaxPool2d(2)
        self.fc = torch.nn.Linear(1408, 10)

    def forward(self, x):
        in_size = x.size(0)
        x = F.relu(self.mp(self.conv1(x)))#conv1 卷积完之后的通道为10，与incep1定义的输入通道相等
        x = self.incep1(x)  #incep1 进行计算
        #incep1计算完成后，输出通道数为 24+16+24+24=88，与conv2层的输入通道数相等
        x = F.relu(self.mp(self.conv2(x)))#conv2输入通道数为上一层的输出通道数88，卷积完之后的通道为20，与incep2定义的输入通道相等
        x = self.incep2(x)  #incep2 进行计算
        x = x.view(in_size, -1)
        x = self.fc(x)

        return x

打印网络格式如下：


使用mnist数据集对编写的网络进行训练，得到的训练结果如下：

LOSS图像与Accuracy图像曲线如下图：

完整实现代码：

# -*- coding: UTF-8 -*-
'''===============================================
@Author ：kidding
@Date   ：2021/2/20 8:56
@File   ：Inception
@IDE    ：PyCharm
=================================================='''
import torch
import numpy as np
from torchvision import transforms
from torchvision import datasets
import torch.nn.functional as F
from torch.utils.data import DataLoader,Dataset
import matplotlib.pyplot as plt
from torchsummary import summary

'''
1、PrePare Dataset
'''
batch_size = 32
#transforms.ToTensor() 将一个PIL图像 转变为一个Tensor，像素值为[0,1]
transform = transforms.Compose([transforms.ToTensor(),
                               transforms.Normalize((0.1307,),(0.3081,))
                               ])
train_dataset = datasets.MNIST(root='./dataset/mnist',train=True,download=False,transform=transform)
train_loader = DataLoader(train_dataset,batch_size=batch_size,shuffle=True)

test_dataset =  datasets.MNIST(root='./dataset/mnist',train=False,download=False,transform=transform)
test_loader = DataLoader(test_dataset,batch_size=batch_size,shuffle=False)

'''
2、Design Model
'''
class Inception(torch.nn.Module):
    def __init__(self,in_channels):
        super(Inception,self).__init__()
        # 定义分支1
        self.branch1x1 = torch.nn.Conv2d(in_channels,16,kernel_size = 1)
        # 定义分支2
        self.branch5x5_1 = torch.nn.Conv2d(in_channels,16,kernel_size = 1)
        self.branch5x5_2 = torch.nn.Conv2d(16,24,kernel_size = 5,padding =2)
        # 定义分支3
        self.branch3x3_1 = torch.nn.Conv2d(in_channels,16,kernel_size=1)
        self.branch3x3_2 = torch.nn.Conv2d(16,24,kernel_size=  3,padding =1)
        self.branch3x3_3 = torch.nn.Conv2d(24,24,kernel_size=  3,padding =1)
        # 定义分支4
        self.branch_pool = torch.nn.Conv2d(in_channels,24,kernel_size = 1)

    def forward(self,x):
        # 计算分支1
        branch1x1 = self.branch1x1(x)
        # 计算分支2
        branch5x5 = self.branch5x5_1(x)
        branch5x5=self.branch5x5_2(branch5x5)
        # 计算分支4
        branch3x3 = self.branch3x3_1(x)
        branch3x3 = self.branch3x3_2(branch3x3)
        branch3x3 = self.branch3x3_3(branch3x3)
        # 计算分支4
        branch_pool = F.avg_pool2d(x, kernel_size=3, stride=1, padding=1)
        branch_pool = self.branch_pool(branch_pool)

        #拼接四个不同分支得到的通道，作为输出
        outputs = [branch1x1, branch5x5, branch3x3, branch_pool]
        return torch.cat(outputs, dim=1)  # b,c,w,h  c对应的是dim=1


class Net(torch.nn.Module):
    def __init__(self):
        super(Net, self).__init__()
        #两个卷积层
        self.conv1 = torch.nn.Conv2d(1, 10, kernel_size=5)
        self.conv2 = torch.nn.Conv2d(88, 20, kernel_size=5)  # 88 = 24x3 + 16

        #两个Inception模块
        self.incep1 = Inception(in_channels=10)  # 输入为10，与conv1 中的10对应
        self.incep2 = Inception(in_channels=20)  # 输入为20，与conv2 中的20对应

        self.mp = torch.nn.MaxPool2d(2)
        self.fc = torch.nn.Linear(1408, 10)

    def forward(self, x):
        in_size = x.size(0)
        x = F.relu(self.mp(self.conv1(x)))#conv1 卷积完之后的通道为10，与incep1定义的输入通道相等
        x = self.incep1(x)  #incep1 进行计算
        #incep1计算完成后，输出通道数为 24+16+24+24=88，与conv2层的输入通道数相等
        x = F.relu(self.mp(self.conv2(x)))#conv2输入通道数为上一层的输出通道数88，卷积完之后的通道为20，与incep2定义的输入通道相等
        x = self.incep2(x)  #incep2 进行计算
        x = x.view(in_size, -1)
        x = self.fc(x)

        return x

#实例化
model = Net()
device = torch.device("cuda" if torch.cuda.is_available() else "cpu")
model.to(device)
#summary(model,input_size=(1,28,28))
'''
3、Construct loss and optimizer
'''
#定义一个损失函数，来计算我们模型输出的值和标准值的差距
criterion = torch.nn.CrossEntropyLoss()
#定义一个优化器，训练模型咋训练的，就靠这个，他会反向的更改相应层的权重
optimizer = torch.optim.SGD(model.parameters(),lr=0.01,momentum = 0.5)

'''
4、Training Cycle : forward-->backward-->update
'''
def train(epoch):
    running_loss = 0.0
    for batch_idx,data in enumerate(train_loader,0): #每次取一个样本
        inputs,target = data
        inputs, target = inputs.to(device), target.to(device)
        optimizer.zero_grad() #优化器清0

        #forward
        outputs = model(inputs)
        loss = criterion(outputs,target)

        # backeard 反向求梯度
        loss.backward()
        # update 更新权重
        optimizer.step()

        running_loss += loss.item() #把损失加起来
        # 每300次输出一下数据
        if batch_idx % 600 == 599:
           # print('[%d,%5d] loss :%3f'%(epoch+1,batch_idx+1,running_loss/300))
            print("Epoch:",epoch+1,"batch_idx:",batch_idx+1,"Loss={:.4f}".format(running_loss/600))
            running_loss = 0.0
    return running_loss/600

def test():
    correct = 0
    total = 0
    with torch.no_grad(): #测试不会计算梯度
        for data in test_loader:
            images,labels = data
            images, labels = images.to(device), labels.to(device)
            outputs = model(images)
            # 我们取概率最大的那个数作为输出
            _,predicted = torch.max(outputs.data,dim=1) #取出最大值的下标   "_,"为占位符，代表每行最大值的下标， dim = 1 列是第0个维度，行是第1个维度
            total += labels.size(0)
            # 计算正确率
            correct += (predicted == labels).sum().item() # 张量之间的比较运算

    #print('Accuracy on Test_set :%d %%'%(100 * correct / total))
    print('Accuracy on test set: %d %% [%d/%d]' % (100 * correct / total, correct, total))
    return correct / total
'''
5、main
'''
if __name__=='__main__':
    epoch_list = []  # 保存epoch
    acc_list = []  # 保存每个epoch的准确率
    Loss_List = []  # 保存每个epoch对应的loss

    for epoch in range(10):
        loss = train(epoch)
        acc = test()

        epoch_list.append(epoch)
        acc_list.append(acc)
        Loss_List.append(loss)

    plt.plot(epoch_list, Loss_List)
    plt.ylabel('Loss')
    plt.xlabel('Epoch')
    plt.grid(ls='--')  # 生成网格
    plt.show()

    plt.plot(epoch_list, acc_list)
    plt.ylabel('accuracy')
    plt.xlabel('epoch')
    plt.grid(ls='--')  # 生成网格
    plt.show()