1.ResNet简介

这个我已经写过一篇文章了感兴趣的去翻一翻

2.ResNet34的简单实现

#!/usr/bin/env python3
# -*- coding: utf-8 -*-
"""
Created on Tue Mar 17 09:31:45 2020

@author: elliot
"""

import torch as t
import torch.nn as nn
from torch.nn import functional as F


class ResidualBlock(nn.Module):
    #显式的继承自nn.Module
    #resnet是卷积的一种
    def __init__(self, inchannel, outchannel, stride=1, shortcut=None):
        #shortcut是直连，resnet和densenet的精髓所在
        #层的定义都在初始化里
        super(ResidualBlock, self).__init__()
        self.left = nn.Sequential(
                nn.Conv2d(inchannel, outchannel, 3, stride, 1, bias=False),
                nn.BatchNorm2d(outchannel),
                nn.ReLU(inplace=True),
                nn.Conv2d(outchannel,outchannel,3,1,1,bias=False),
                nn.BatchNorm2d(outchannel))
        self.right = shortcut
    
    def forward(self,x):
        out = self.left(x)
        residual = x if self.right is None else self.right(x)
        out += residual
        return F.relu(out)
    

class ResNet(nn.Module):
    #包括34，50，101等多种结构，可以按需实现，这里是Resnet34
    def __init__(self, num_classes=1000):
        super(ResNet,self).__init__()
        self.pre = nn.Sequential(nn.Conv2d(3,64,7,2,3,bias=False),
                                 nn.BatchNorm2d(64),#这个64是指feature_num
                                 nn.ReLU(inplace=True),
                                 nn.MaxPool2d(3,2,1) )
        self.layer1 = self._make_layer(64, 128, 3)
        self.layer2 = self._make_layer(128, 256, 4, stride=2)
        self.layer3 = self._make_layer(256, 512, 6, stride=2)
        self.layer4 = self._make_layer(512, 512, 3, stride=2)
        self.fc = nn.Linear(512, num_classes)

    def _make_layer(self, inchannel, outchannel, block_num, stride=1):
        short_cut = nn.Sequential(
                nn.Conv2d(inchannel,outchannel,1,stride,bias=False),
                nn.BatchNorm2d(outchannel)
                )
        layers = []
        layers.append(ResidualBlock(inchannel,outchannel,stride,short_cut))
        for i in range(1, block_num):
            layers.append(ResidualBlock(outchannel,outchannel))#输入和输出要一致
        return nn.Sequential(*layers)
    
    def forward(self, x):
        x = self.pre(x)
        x = self.layer1(x)
        x = self.layer2(x)
        x = self.layer3(x)
        x = self.layer4(x)
        x = F.avg_pool2d(x,7)#注意F和原生的区别
        x = x.view(x.size(0), -1)
        return self.fc(x)
            
    
if __name__ == '__main__':
    net = ResNet()
    print(net)
    input = t.autograd.Variable(t.randn(1,3,224,224))
    o = net(input)
    print(o)