GoogleNet

GoogleNet

一.手写GoogleNet架构

 

import torch
from d2l import torch as d2l
from torch import nn
from torch.nn import functional as F
#1.Inception块
class Inception(nn.Module):
    #一个Inception块包含四条路径
    #in_chanels输入通道数，c1~c4表示每条路径上的输出通道数量，如果该路径上有多个卷积层，那输出通道应该是列表
    def __init__(self,in_chanels,c1,c2,c3,c4,**kwargs):
        super(Inception,self).__init__(**kwargs)
        #路径1:1*1卷积层
        self.p1_1=nn.Conv2d(in_chanels,c1,1)
        #路径2:1*1卷积->3*3卷积，padding=1
        self.p2_1=nn.Conv2d(in_chanels,c2[0],1)
        self.p2_2=nn.Conv2d(c2[0],c2[1],3,padding=1)
        #路径3：1*1卷积->5*5卷积，padding=2
        self.p3_1=nn.Conv2d(in_chanels,c3[0],1)
        self.p3_2=nn.Conv2d(c3[0],c3[1],5,padding=2)
        #路径4:3*3最大池化层padding=1，stride=1->1*1卷积
        self.p4_1=nn.MaxPool2d(3,padding=1,stride=1)
        self.p4_2=nn.Conv2d(in_chanels,c4,kernel_size=1)
    def forward(self,x):
        p1=F.relu(self.p1_1(x))
        p2=F.relu(self.p2_2(F.relu(self.p2_1(x))))
        p3=F.relu(self.p3_2(F.relu(self.p3_1(x))))
        p4=F.relu(self.p4_2(self.p4_1(x)))
        return torch.cat((p1,p2,p3,p4),dim=1)
#每一个stage
#第一个stage
b1=nn.Sequential(
    nn.Conv2d(1,64,7,2,3),
    nn.ReLU(),
    nn.MaxPool2d(3,2,1)
)
#第2个stage
b2=nn.Sequential(
    nn.Conv2d(64,64,1),
    nn.ReLU(),
    nn.Conv2d(64,192,3,padding=1),
    nn.ReLU(),
    nn.MaxPool2d(3,2,1)
)
#Inception块数自己定
#第3个stage:2个Inception块和最大池化层
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))
#stage4:5个Inception块，一个最大池化层
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))
#stage5:2个Inception块，全局平均池化层AdaptiveAvgPool2d
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())
#最终输出通道数为10因为想做数字0~9的分类
net = nn.Sequential(b1, b2, b3, b4, b5, nn.Linear(1024, 10))
X=torch.rand((1,1,96,96))
for layer in net:
    X=layer(X)
    print(layer.__class__.__name__,X.shape)

二.训练GoogleNet

import torch
from d2l import torch as d2l
from torch import nn
from torch.nn import functional as F
#1.Inception块
class Inception(nn.Module):
    #一个Inception块包含四条路径
    #in_chanels输入通道数，c1~c4表示每条路径上的输出通道数量，如果该路径上有多个卷积层，那输出通道应该是列表
    def __init__(self,in_chanels,c1,c2,c3,c4,**kwargs):
        super(Inception,self).__init__(**kwargs)
        #路径1:1*1卷积层
        self.p1_1=nn.Conv2d(in_chanels,c1,1)
        #路径2:1*1卷积->3*3卷积，padding=1
        self.p2_1=nn.Conv2d(in_chanels,c2[0],1)
        self.p2_2=nn.Conv2d(c2[0],c2[1],3,padding=1)
        #路径3：1*1卷积->5*5卷积，padding=2
        self.p3_1=nn.Conv2d(in_chanels,c3[0],1)
        self.p3_2=nn.Conv2d(c3[0],c3[1],5,padding=2)
        #路径4:3*3最大池化层padding=1，stride=1->1*1卷积
        self.p4_1=nn.MaxPool2d(3,padding=1,stride=1)
        self.p4_2=nn.Conv2d(in_chanels,c4,kernel_size=1)
    def forward(self,x):
        p1=F.relu(self.p1_1(x))
        p2=F.relu(self.p2_2(F.relu(self.p2_1(x))))
        p3=F.relu(self.p3_2(F.relu(self.p3_1(x))))
        p4=F.relu(self.p4_2(self.p4_1(x)))
        return torch.cat((p1,p2,p3,p4),dim=1)
#每一个stage
#第一个stage
b1=nn.Sequential(
    nn.Conv2d(1,64,7,2,3),
    nn.ReLU(),
    nn.MaxPool2d(3,2,1)
)
#第2个stage
b2=nn.Sequential(
    nn.Conv2d(64,64,1),
    nn.ReLU(),
    nn.Conv2d(64,192,3,padding=1),
    nn.ReLU(),
    nn.MaxPool2d(3,2,1)
)
#Inception块数自己定
#第3个stage:2个Inception块和最大池化层
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))
#stage4:5个Inception块，一个最大池化层
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))
#stage5:2个Inception块，全局平均池化层AdaptiveAvgPool2d
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())
#最终输出通道数为10因为想做数字0~9的分类
net = nn.Sequential(b1, b2, b3, b4, b5, nn.Linear(1024, 10))
train_iter,test_iter=d2l.load_data_fashion_mnist(batch_size=128,resize=96)
lr=0.1
nums_echos=10
d2l.train_ch6(net,train_iter,test_iter,nums_echos,lr,d2l.try_gpu())