AI 助理
备案控制台
开发者社区 云计算 文章 正文

ResNet系列网络结构介绍

2022-06-09 739
版权
版权声明：
本文内容由阿里云实名注册用户自发贡献，版权归原作者所有，阿里云开发者社区不拥有其著作权，亦不承担相应法律责任。具体规则请查看《 阿里云开发者社区用户服务协议》和 《阿里云开发者社区知识产权保护指引》。如果您发现本社区中有涉嫌抄袭的内容，填写 侵权投诉表单进行举报，一经查实，本社区将立刻删除涉嫌侵权内容。
简介： ResNet系列网络结构介绍

前言

这里将着重介绍网络的结构构成，不涉及原理讲解。



一.ResNet18结构

层数: 71


连接数: 78


结构M

%添加层分支
%将网络分支添加到层次图中。每个分支均为一个线性层组。
tempLayers = [
    imageInputLayer([224 224 3],"Name","data","Normalization","zscore")
    convolution2dLayer([7 7],64,"Name","conv1","BiasLearnRateFactor",0,"Padding",[3 3 3 3],"Stride",[2 2])
    batchNormalizationLayer("Name","bn_conv1")
    reluLayer("Name","conv1_relu")
    maxPooling2dLayer([3 3],"Name","pool1","Padding",[1 1 1 1],"Stride",[2 2])];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([3 3],64,"Name","res2a_branch2a","BiasLearnRateFactor",0,"Padding",[1 1 1 1])
    batchNormalizationLayer("Name","bn2a_branch2a")
    reluLayer("Name","res2a_branch2a_relu")
    convolution2dLayer([3 3],64,"Name","res2a_branch2b","BiasLearnRateFactor",0,"Padding",[1 1 1 1])
    batchNormalizationLayer("Name","bn2a_branch2b")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","res2a")
    reluLayer("Name","res2a_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([3 3],64,"Name","res2b_branch2a","BiasLearnRateFactor",0,"Padding",[1 1 1 1])
    batchNormalizationLayer("Name","bn2b_branch2a")
    reluLayer("Name","res2b_branch2a_relu")
    convolution2dLayer([3 3],64,"Name","res2b_branch2b","BiasLearnRateFactor",0,"Padding",[1 1 1 1])
    batchNormalizationLayer("Name","bn2b_branch2b")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","res2b")
    reluLayer("Name","res2b_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([3 3],128,"Name","res3a_branch2a","BiasLearnRateFactor",0,"Padding",[1 1 1 1],"Stride",[2 2])
    batchNormalizationLayer("Name","bn3a_branch2a")
    reluLayer("Name","res3a_branch2a_relu")
    convolution2dLayer([3 3],128,"Name","res3a_branch2b","BiasLearnRateFactor",0,"Padding",[1 1 1 1])
    batchNormalizationLayer("Name","bn3a_branch2b")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],128,"Name","res3a_branch1","BiasLearnRateFactor",0,"Stride",[2 2])
    batchNormalizationLayer("Name","bn3a_branch1")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","res3a")
    reluLayer("Name","res3a_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([3 3],128,"Name","res3b_branch2a","BiasLearnRateFactor",0,"Padding",[1 1 1 1])
    batchNormalizationLayer("Name","bn3b_branch2a")
    reluLayer("Name","res3b_branch2a_relu")
    convolution2dLayer([3 3],128,"Name","res3b_branch2b","BiasLearnRateFactor",0,"Padding",[1 1 1 1])
    batchNormalizationLayer("Name","bn3b_branch2b")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","res3b")
    reluLayer("Name","res3b_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([3 3],256,"Name","res4a_branch2a","BiasLearnRateFactor",0,"Padding",[1 1 1 1],"Stride",[2 2])
    batchNormalizationLayer("Name","bn4a_branch2a")
    reluLayer("Name","res4a_branch2a_relu")
    convolution2dLayer([3 3],256,"Name","res4a_branch2b","BiasLearnRateFactor",0,"Padding",[1 1 1 1])
    batchNormalizationLayer("Name","bn4a_branch2b")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],256,"Name","res4a_branch1","BiasLearnRateFactor",0,"Stride",[2 2])
    batchNormalizationLayer("Name","bn4a_branch1")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","res4a")
    reluLayer("Name","res4a_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([3 3],256,"Name","res4b_branch2a","BiasLearnRateFactor",0,"Padding",[1 1 1 1])
    batchNormalizationLayer("Name","bn4b_branch2a")
    reluLayer("Name","res4b_branch2a_relu")
    convolution2dLayer([3 3],256,"Name","res4b_branch2b","BiasLearnRateFactor",0,"Padding",[1 1 1 1])
    batchNormalizationLayer("Name","bn4b_branch2b")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","res4b")
    reluLayer("Name","res4b_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],512,"Name","res5a_branch1","BiasLearnRateFactor",0,"Stride",[2 2])
    batchNormalizationLayer("Name","bn5a_branch1")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([3 3],512,"Name","res5a_branch2a","BiasLearnRateFactor",0,"Padding",[1 1 1 1],"Stride",[2 2])
    batchNormalizationLayer("Name","bn5a_branch2a")
    reluLayer("Name","res5a_branch2a_relu")
    convolution2dLayer([3 3],512,"Name","res5a_branch2b","BiasLearnRateFactor",0,"Padding",[1 1 1 1])
    batchNormalizationLayer("Name","bn5a_branch2b")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","res5a")
    reluLayer("Name","res5a_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([3 3],512,"Name","res5b_branch2a","BiasLearnRateFactor",0,"Padding",[1 1 1 1])
    batchNormalizationLayer("Name","bn5b_branch2a")
    reluLayer("Name","res5b_branch2a_relu")
    convolution2dLayer([3 3],512,"Name","res5b_branch2b","BiasLearnRateFactor",0,"Padding",[1 1 1 1])
    batchNormalizationLayer("Name","bn5b_branch2b")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","res5b")
    reluLayer("Name","res5b_relu")
    globalAveragePooling2dLayer("Name","pool5")
    fullyConnectedLayer(1000,"Name","fc1000")
    softmaxLayer("Name","prob")
    classificationLayer("Name","ClassificationLayer_predictions")];
lgraph = addLayers(lgraph,tempLayers);
% 清理辅助函数变量
clear tempLayers;
连接层分支
连接网络的所有分支以创建网络图。
lgraph = connectLayers(lgraph,"pool1","res2a_branch2a");
lgraph = connectLayers(lgraph,"pool1","res2a/in2");
lgraph = connectLayers(lgraph,"bn2a_branch2b","res2a/in1");
lgraph = connectLayers(lgraph,"res2a_relu","res2b_branch2a");
lgraph = connectLayers(lgraph,"res2a_relu","res2b/in2");
lgraph = connectLayers(lgraph,"bn2b_branch2b","res2b/in1");
lgraph = connectLayers(lgraph,"res2b_relu","res3a_branch2a");
lgraph = connectLayers(lgraph,"res2b_relu","res3a_branch1");
lgraph = connectLayers(lgraph,"bn3a_branch1","res3a/in2");
lgraph = connectLayers(lgraph,"bn3a_branch2b","res3a/in1");
lgraph = connectLayers(lgraph,"res3a_relu","res3b_branch2a");
lgraph = connectLayers(lgraph,"res3a_relu","res3b/in2");
lgraph = connectLayers(lgraph,"bn3b_branch2b","res3b/in1");
lgraph = connectLayers(lgraph,"res3b_relu","res4a_branch2a");
lgraph = connectLayers(lgraph,"res3b_relu","res4a_branch1");
lgraph = connectLayers(lgraph,"bn4a_branch1","res4a/in2");
lgraph = connectLayers(lgraph,"bn4a_branch2b","res4a/in1");
lgraph = connectLayers(lgraph,"res4a_relu","res4b_branch2a");
lgraph = connectLayers(lgraph,"res4a_relu","res4b/in2");
lgraph = connectLayers(lgraph,"bn4b_branch2b","res4b/in1");
lgraph = connectLayers(lgraph,"res4b_relu","res5a_branch1");
lgraph = connectLayers(lgraph,"res4b_relu","res5a_branch2a");
lgraph = connectLayers(lgraph,"bn5a_branch1","res5a/in2");
lgraph = connectLayers(lgraph,"bn5a_branch2b","res5a/in1");
lgraph = connectLayers(lgraph,"res5a_relu","res5b_branch2a");
lgraph = connectLayers(lgraph,"res5a_relu","res5b/in2");
lgraph = connectLayers(lgraph,"bn5b_branch2b","res5b/in1");
%绘制层
plot(lgraph);



结构图内部参数计


image.pngimage.pngimage.png


算图

image.png

image.png

image.png


二.ResNet 50结构

层数: 177


连接数: 192


结构m:

%创建层次图
%创建层次图变量以包含网络层。
lgraph = layerGraph();
%添加层分支
%将网络分支添加到层次图中。每个分支均为一个线性层组。
tempLayers = [
    imageInputLayer([224 224 3],"Name","input_1")
    convolution2dLayer([7 7],64,"Name","conv1","Padding",[3 3 3 3],"Stride",[2 2])
    batchNormalizationLayer("Name","bn_conv1","Epsilon",0.001)
    reluLayer("Name","activation_1_relu")
    maxPooling2dLayer([3 3],"Name","max_pooling2d_1","Padding",[1 1 1 1],"Stride",[2 2])];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],256,"Name","res2a_branch1","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn2a_branch1","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],64,"Name","res2a_branch2a","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn2a_branch2a","Epsilon",0.001)
    reluLayer("Name","activation_2_relu")
    convolution2dLayer([3 3],64,"Name","res2a_branch2b","BiasLearnRateFactor",0,"Padding","same")
    batchNormalizationLayer("Name","bn2a_branch2b","Epsilon",0.001)
    reluLayer("Name","activation_3_relu")
    convolution2dLayer([1 1],256,"Name","res2a_branch2c","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn2a_branch2c","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","add_1")
    reluLayer("Name","activation_4_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],64,"Name","res2b_branch2a","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn2b_branch2a","Epsilon",0.001)
    reluLayer("Name","activation_5_relu")
    convolution2dLayer([3 3],64,"Name","res2b_branch2b","BiasLearnRateFactor",0,"Padding","same")
    batchNormalizationLayer("Name","bn2b_branch2b","Epsilon",0.001)
    reluLayer("Name","activation_6_relu")
    convolution2dLayer([1 1],256,"Name","res2b_branch2c","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn2b_branch2c","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","add_2")
    reluLayer("Name","activation_7_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],64,"Name","res2c_branch2a","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn2c_branch2a","Epsilon",0.001)
    reluLayer("Name","activation_8_relu")
    convolution2dLayer([3 3],64,"Name","res2c_branch2b","BiasLearnRateFactor",0,"Padding","same")
    batchNormalizationLayer("Name","bn2c_branch2b","Epsilon",0.001)
    reluLayer("Name","activation_9_relu")
    convolution2dLayer([1 1],256,"Name","res2c_branch2c","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn2c_branch2c","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","add_3")
    reluLayer("Name","activation_10_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],128,"Name","res3a_branch2a","BiasLearnRateFactor",0,"Stride",[2 2])
    batchNormalizationLayer("Name","bn3a_branch2a","Epsilon",0.001)
    reluLayer("Name","activation_11_relu")
    convolution2dLayer([3 3],128,"Name","res3a_branch2b","BiasLearnRateFactor",0,"Padding","same")
    batchNormalizationLayer("Name","bn3a_branch2b","Epsilon",0.001)
    reluLayer("Name","activation_12_relu")
    convolution2dLayer([1 1],512,"Name","res3a_branch2c","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn3a_branch2c","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],512,"Name","res3a_branch1","BiasLearnRateFactor",0,"Stride",[2 2])
    batchNormalizationLayer("Name","bn3a_branch1","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","add_4")
    reluLayer("Name","activation_13_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],128,"Name","res3b_branch2a","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn3b_branch2a","Epsilon",0.001)
    reluLayer("Name","activation_14_relu")
    convolution2dLayer([3 3],128,"Name","res3b_branch2b","BiasLearnRateFactor",0,"Padding","same")
    batchNormalizationLayer("Name","bn3b_branch2b","Epsilon",0.001)
    reluLayer("Name","activation_15_relu")
    convolution2dLayer([1 1],512,"Name","res3b_branch2c","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn3b_branch2c","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","add_5")
    reluLayer("Name","activation_16_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],128,"Name","res3c_branch2a","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn3c_branch2a","Epsilon",0.001)
    reluLayer("Name","activation_17_relu")
    convolution2dLayer([3 3],128,"Name","res3c_branch2b","BiasLearnRateFactor",0,"Padding","same")
    batchNormalizationLayer("Name","bn3c_branch2b","Epsilon",0.001)
    reluLayer("Name","activation_18_relu")
    convolution2dLayer([1 1],512,"Name","res3c_branch2c","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn3c_branch2c","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","add_6")
    reluLayer("Name","activation_19_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],128,"Name","res3d_branch2a","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn3d_branch2a","Epsilon",0.001)
    reluLayer("Name","activation_20_relu")
    convolution2dLayer([3 3],128,"Name","res3d_branch2b","BiasLearnRateFactor",0,"Padding","same")
    batchNormalizationLayer("Name","bn3d_branch2b","Epsilon",0.001)
    reluLayer("Name","activation_21_relu")
    convolution2dLayer([1 1],512,"Name","res3d_branch2c","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn3d_branch2c","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","add_7")
    reluLayer("Name","activation_22_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],1024,"Name","res4a_branch1","BiasLearnRateFactor",0,"Stride",[2 2])
    batchNormalizationLayer("Name","bn4a_branch1","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],256,"Name","res4a_branch2a","BiasLearnRateFactor",0,"Stride",[2 2])
    batchNormalizationLayer("Name","bn4a_branch2a","Epsilon",0.001)
    reluLayer("Name","activation_23_relu")
    convolution2dLayer([3 3],256,"Name","res4a_branch2b","BiasLearnRateFactor",0,"Padding","same")
    batchNormalizationLayer("Name","bn4a_branch2b","Epsilon",0.001)
    reluLayer("Name","activation_24_relu")
    convolution2dLayer([1 1],1024,"Name","res4a_branch2c","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn4a_branch2c","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","add_8")
    reluLayer("Name","activation_25_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],256,"Name","res4b_branch2a","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn4b_branch2a","Epsilon",0.001)
    reluLayer("Name","activation_26_relu")
    convolution2dLayer([3 3],256,"Name","res4b_branch2b","BiasLearnRateFactor",0,"Padding","same")
    batchNormalizationLayer("Name","bn4b_branch2b","Epsilon",0.001)
    reluLayer("Name","activation_27_relu")
    convolution2dLayer([1 1],1024,"Name","res4b_branch2c","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn4b_branch2c","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","add_9")
    reluLayer("Name","activation_28_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],256,"Name","res4c_branch2a","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn4c_branch2a","Epsilon",0.001)
    reluLayer("Name","activation_29_relu")
    convolution2dLayer([3 3],256,"Name","res4c_branch2b","BiasLearnRateFactor",0,"Padding","same")
    batchNormalizationLayer("Name","bn4c_branch2b","Epsilon",0.001)
    reluLayer("Name","activation_30_relu")
    convolution2dLayer([1 1],1024,"Name","res4c_branch2c","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn4c_branch2c","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","add_10")
    reluLayer("Name","activation_31_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],256,"Name","res4d_branch2a","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn4d_branch2a","Epsilon",0.001)
    reluLayer("Name","activation_32_relu")
    convolution2dLayer([3 3],256,"Name","res4d_branch2b","BiasLearnRateFactor",0,"Padding","same")
    batchNormalizationLayer("Name","bn4d_branch2b","Epsilon",0.001)
    reluLayer("Name","activation_33_relu")
    convolution2dLayer([1 1],1024,"Name","res4d_branch2c","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn4d_branch2c","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","add_11")
    reluLayer("Name","activation_34_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],256,"Name","res4e_branch2a","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn4e_branch2a","Epsilon",0.001)
    reluLayer("Name","activation_35_relu")
    convolution2dLayer([3 3],256,"Name","res4e_branch2b","BiasLearnRateFactor",0,"Padding","same")
    batchNormalizationLayer("Name","bn4e_branch2b","Epsilon",0.001)
    reluLayer("Name","activation_36_relu")
    convolution2dLayer([1 1],1024,"Name","res4e_branch2c","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn4e_branch2c","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","add_12")
    reluLayer("Name","activation_37_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],256,"Name","res4f_branch2a","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn4f_branch2a","Epsilon",0.001)
    reluLayer("Name","activation_38_relu")
    convolution2dLayer([3 3],256,"Name","res4f_branch2b","BiasLearnRateFactor",0,"Padding","same")
    batchNormalizationLayer("Name","bn4f_branch2b","Epsilon",0.001)
    reluLayer("Name","activation_39_relu")
    convolution2dLayer([1 1],1024,"Name","res4f_branch2c","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn4f_branch2c","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","add_13")
    reluLayer("Name","activation_40_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],2048,"Name","res5a_branch1","BiasLearnRateFactor",0,"Stride",[2 2])
    batchNormalizationLayer("Name","bn5a_branch1","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],512,"Name","res5a_branch2a","BiasLearnRateFactor",0,"Stride",[2 2])
    batchNormalizationLayer("Name","bn5a_branch2a","Epsilon",0.001)
    reluLayer("Name","activation_41_relu")
    convolution2dLayer([3 3],512,"Name","res5a_branch2b","BiasLearnRateFactor",0,"Padding","same")
    batchNormalizationLayer("Name","bn5a_branch2b","Epsilon",0.001)
    reluLayer("Name","activation_42_relu")
    convolution2dLayer([1 1],2048,"Name","res5a_branch2c","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn5a_branch2c","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","add_14")
    reluLayer("Name","activation_43_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],512,"Name","res5b_branch2a","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn5b_branch2a","Epsilon",0.001)
    reluLayer("Name","activation_44_relu")
    convolution2dLayer([3 3],512,"Name","res5b_branch2b","BiasLearnRateFactor",0,"Padding","same")
    batchNormalizationLayer("Name","bn5b_branch2b","Epsilon",0.001)
    reluLayer("Name","activation_45_relu")
    convolution2dLayer([1 1],2048,"Name","res5b_branch2c","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn5b_branch2c","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","add_15")
    reluLayer("Name","activation_46_relu")];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    convolution2dLayer([1 1],512,"Name","res5c_branch2a","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn5c_branch2a","Epsilon",0.001)
    reluLayer("Name","activation_47_relu")
    convolution2dLayer([3 3],512,"Name","res5c_branch2b","BiasLearnRateFactor",0,"Padding","same")
    batchNormalizationLayer("Name","bn5c_branch2b","Epsilon",0.001)
    reluLayer("Name","activation_48_relu")
    convolution2dLayer([1 1],2048,"Name","res5c_branch2c","BiasLearnRateFactor",0)
    batchNormalizationLayer("Name","bn5c_branch2c","Epsilon",0.001)];
lgraph = addLayers(lgraph,tempLayers);
tempLayers = [
    additionLayer(2,"Name","add_16")
    reluLayer("Name","activation_49_relu")
    globalAveragePooling2dLayer("Name","avg_pool")
    fullyConnectedLayer(1000,"Name","fc1000","BiasLearnRateFactor",0)
    softmaxLayer("Name","fc1000_softmax")
    classificationLayer("Name","ClassificationLayer_fc1000")];
lgraph = addLayers(lgraph,tempLayers);
% 清理辅助函数变量
clear tempLayers;
%连接层分支
%连接网络的所有分支以创建网络图。
lgraph = connectLayers(lgraph,"max_pooling2d_1","res2a_branch1");
lgraph = connectLayers(lgraph,"max_pooling2d_1","res2a_branch2a");
lgraph = connectLayers(lgraph,"bn2a_branch1","add_1/in2");
lgraph = connectLayers(lgraph,"bn2a_branch2c","add_1/in1");
lgraph = connectLayers(lgraph,"activation_4_relu","res2b_branch2a");
lgraph = connectLayers(lgraph,"activation_4_relu","add_2/in2");
lgraph = connectLayers(lgraph,"bn2b_branch2c","add_2/in1");
lgraph = connectLayers(lgraph,"activation_7_relu","res2c_branch2a");
lgraph = connectLayers(lgraph,"activation_7_relu","add_3/in2");
lgraph = connectLayers(lgraph,"bn2c_branch2c","add_3/in1");
lgraph = connectLayers(lgraph,"activation_10_relu","res3a_branch2a");
lgraph = connectLayers(lgraph,"activation_10_relu","res3a_branch1");
lgraph = connectLayers(lgraph,"bn3a_branch2c","add_4/in1");
lgraph = connectLayers(lgraph,"bn3a_branch1","add_4/in2");
lgraph = connectLayers(lgraph,"activation_13_relu","res3b_branch2a");
lgraph = connectLayers(lgraph,"activation_13_relu","add_5/in2");
lgraph = connectLayers(lgraph,"bn3b_branch2c","add_5/in1");
lgraph = connectLayers(lgraph,"activation_16_relu","res3c_branch2a");
lgraph = connectLayers(lgraph,"activation_16_relu","add_6/in2");
lgraph = connectLayers(lgraph,"bn3c_branch2c","add_6/in1");
lgraph = connectLayers(lgraph,"activation_19_relu","res3d_branch2a");
lgraph = connectLayers(lgraph,"activation_19_relu","add_7/in2");
lgraph = connectLayers(lgraph,"bn3d_branch2c","add_7/in1");
lgraph = connectLayers(lgraph,"activation_22_relu","res4a_branch1");
lgraph = connectLayers(lgraph,"activation_22_relu","res4a_branch2a");
lgraph = connectLayers(lgraph,"bn4a_branch1","add_8/in2");
lgraph = connectLayers(lgraph,"bn4a_branch2c","add_8/in1");
lgraph = connectLayers(lgraph,"activation_25_relu","res4b_branch2a");
lgraph = connectLayers(lgraph,"activation_25_relu","add_9/in2");
lgraph = connectLayers(lgraph,"bn4b_branch2c","add_9/in1");
lgraph = connectLayers(lgraph,"activation_28_relu","res4c_branch2a");
lgraph = connectLayers(lgraph,"activation_28_relu","add_10/in2");
lgraph = connectLayers(lgraph,"bn4c_branch2c","add_10/in1");
lgraph = connectLayers(lgraph,"activation_31_relu","res4d_branch2a");
lgraph = connectLayers(lgraph,"activation_31_relu","add_11/in2");
lgraph = connectLayers(lgraph,"bn4d_branch2c","add_11/in1");
lgraph = connectLayers(lgraph,"activation_34_relu","res4e_branch2a");
lgraph = connectLayers(lgraph,"activation_34_relu","add_12/in2");
lgraph = connectLayers(lgraph,"bn4e_branch2c","add_12/in1");
lgraph = connectLayers(lgraph,"activation_37_relu","res4f_branch2a");
lgraph = connectLayers(lgraph,"activation_37_relu","add_13/in2");
lgraph = connectLayers(lgraph,"bn4f_branch2c","add_13/in1");
lgraph = connectLayers(lgraph,"activation_40_relu","res5a_branch1");
lgraph = connectLayers(lgraph,"activation_40_relu","res5a_branch2a");
lgraph = connectLayers(lgraph,"bn5a_branch1","add_14/in2");
lgraph = connectLayers(lgraph,"bn5a_branch2c","add_14/in1");
lgraph = connectLayers(lgraph,"activation_43_relu","res5b_branch2a");
lgraph = connectLayers(lgraph,"activation_43_relu","add_15/in2");
lgraph = connectLayers(lgraph,"bn5b_branch2c","add_15/in1");
lgraph = connectLayers(lgraph,"activation_46_relu","res5c_branch2a");
lgraph = connectLayers(lgraph,"activation_46_relu","add_16/in2");
lgraph = connectLayers(lgraph,"bn5c_branch2c","add_16/in1");
%绘制层
plot(lgraph);


结构图内部参数计算图


image.png

image.png

image.png

image.png

image.png


image.png

image.pngimage.png

image.png

image.pngimage.png

image.png



关键词：
网络结构
resnet网络结构
ResNet结构
小小杨树1
目录
相关文章
wljslmz
|
2月前
|
机器学习/深度学习 资源调度 自然语言处理
不同类型的循环神经网络结构
【8月更文挑战第16天】
wljslmz
33 0
源码星辰
|
1月前
|
编解码 人工智能 文件存储
卷积神经网络架构：EfficientNet结构的特点
EfficientNet是一种高效的卷积神经网络架构，它通过系统化的方法来提升模型的性能和效率。
源码星辰
31 1
BetterBench
|
2月前
|
机器学习/深度学习 算法 文件存储
【博士每天一篇文献-算法】 PNN网络启发的神经网络结构搜索算法Progressive neural architecture search
本文提出了一种名为渐进式神经架构搜索（Progressive Neural Architecture Search, PNAS）的方法，它使用顺序模型优化策略和替代模型来逐步搜索并优化卷积神经网络结构，从而提高了搜索效率并减少了训练成本。
BetterBench
40 9
LDG_AGI
|
2月前
|
机器学习/深度学习 人工智能 PyTorch
AI智能体研发之路-模型篇（五）：pytorch vs tensorflow框架DNN网络结构源码级对比
AI智能体研发之路-模型篇（五）：pytorch vs tensorflow框架DNN网络结构源码级对比
LDG_AGI
65 1
楠竹11
|
3月前
|
机器学习/深度学习 自然语言处理
像生物网络一样生长，具备结构可塑性的自组织神经网络来了
【7月更文挑战第24天】Sebastian Risi团队发布的arXiv论文探讨了一种模仿生物神经网络生长与适应特性的新型神经网络。LNDP利用结构可塑性和经验依赖学习，能根据活动与奖励动态调整连接，展现自我组织能力。通过基于图变换器的机制，LNDP支持突触动态增删，预先通过可学习随机过程驱动网络发育。实验在Cartpole等任务中验证了LNDP的有效性，尤其在需快速适应的场景下。然而，LNDP在复杂环境下的可扩展性及训练优化仍面临挑战，且其在大规模网络和图像分类等领域的应用尚待探索
楠竹11
75 20
BetterBench
|
2月前
|
机器学习/深度学习 Linux TensorFlow
【Tensorflow+keras】用代码给神经网络结构绘图
文章提供了使用TensorFlow和Keras来绘制神经网络结构图的方法，并给出了具体的代码示例。
BetterBench
42 0
游客wkzdk2qklv76k
|
2月前
|
机器学习/深度学习 API 数据处理
《零基础实践深度学习》2.4手写数字识别之网络结构
这篇文章介绍了手写数字识别任务中网络结构设计的优化，比较了多层全连接神经网络和卷积神经网络两种模型结构，并展示了使用PaddlePaddle框架实现这些网络结构，训练并观察它们在MNIST数据集上的表现。
游客wkzdk2qklv76k
19 0
源码星辰
|
3月前
|
机器学习/深度学习 自然语言处理 算法
深度学习的关键概念和网络结构
度学习是人工智能和机器学习的一个重要分支，它通过模拟人脑神经元的工作方式来处理复杂的模式识别和数据分析任务。深度学习已经在许多领域取得了显著的成果，如图像识别、语音识别和自然语言处理。
源码星辰
45 1
小空门123-30335
|
3月前
|
机器学习/深度学习 PyTorch 算法框架/工具
图神经网络是一类用于处理图结构数据的神经网络。与传统的深度学习模型（如卷积神经网络CNN和循环神经网络RNN）不同，
图神经网络是一类用于处理图结构数据的神经网络。与传统的深度学习模型（如卷积神经网络CNN和循环神经网络RNN）不同，
小空门123-30335
55 9
众所周知
|
3月前
|
存储 算法 Python
Python图论实战：从零基础到精通DFS与BFS遍历，轻松玩转复杂网络结构
【7月更文挑战第11天】图论在数据科学中扮演关键角色，用于解决复杂网络问题。Python因其易用性和库支持成为实现图算法的首选。本文通过问答形式介绍DFS和BFS，图是节点和边的数据结构，遍历用于搜索和分析。Python中图可表示为邻接表，DFS用递归遍历，BFS借助队列。DFS适用于深度探索，BFS则用于最短路径。提供的代码示例帮助理解如何在Python中应用这两种遍历算法。开始探索图论，解锁更多技术可能！
众所周知
114 6

热门文章

最新文章

  • 1
    倚天产品介绍|倚天性能优化—YCL AI计算库在resnet50上的优化
  • 2
    深度学习第19讲：CNN经典论文研读之残差网络ResNet及其keras实现
  • 3
    图像分类经典神经网络大总结（AlexNet、VGG 、GoogLeNet 、ResNet、 DenseNet、SENet、ResNeXt ）
  • 4
    手撕Resnet卷积神经网络-pytorch-详细注释版（可以直接替换自己数据集）-直接放置自己的数据集就能直接跑。跑的代码有问题的可以在评论区指出，看到了会回复。训练代码和预测代码均有。
  • 5
    【论文泛读】 ResNeXt：深度神经网络的聚合残差变换（ResNet的改进，提出了一种新的维度）
  • 6
    倚天性能优化—YCL AI计算库在resnet50上的优化
  • 7
    ResNet实战：tensorflow2.0以上版本，使用ResNet50实现图像分类任务
  • 8
    改进UNet | 透过UCTransNet分析ResNet+UNet是不是真的有效？（一）
  • 9
    Resnet图像识别入门——池化层
  • 10
    深度学习经典网络模型汇总——LeNet、AlexNet、ZFNet、VGGNet、GoogleNet、ResNet【对卷积池化后特征图变化有详细讲解】
  • 1
    m基于Yolov2深度学习网络的螺丝检测系统matlab仿真,带GUI界面
    51
  • 2
    基于深度学习网络的鞋子种类识别matlab仿真
    475
  • 3
    【C 言专栏】C 语言实现网络通信程序
    161
  • 4
    【AI 生成式】生成对抗网络 （GAN） 的概念
    87
  • 5
    【AI 场景】如何应用人工智能来增强企业网络的网络安全？
    100
  • 6
    智能化运维：AIOps在未来网络管理中的应用与挑战
    279
  • 7
    网络安全与信息安全：防御前线的技术与意识
    40
  • 8
    BiTCN：基于卷积网络的多元时间序列预测
    211
  • 9
    网络防御前线：洞悉网络安全漏洞与加固信息防线
    68
  • 10
    网络防御的三重奏：漏洞管理、加密技术与安全意识提升
    45

    • 相关课程

    更多
  • TCP/IP 网络基础
  • 云计算工程师解析与实战-网络专家篇（体验版）
  • 云网络白皮书-阿里云网络系列课
  • 企业上云攻略-阿里云网络产品应用系列教程
  • Linux网络进阶 - TCP/IP协议及OSI七层模型
  • 网络管理者必知-2分钟了解新出台的《网络安全法》

    • 相关电子书

    更多
  • Session：弹性、高可用、可观测的应用交付网络产品更新
  • Keynote：阿里云网络领域关键创新技术解读
  • 智慧光网络 关键技术应用实践和未来演进

    • 相关实验场景

    更多
  • 如何快速训练大模型
  • 容器的网络入门
  • 容器的自定义网络
  • 容器的共享网络模型
  • 使用阿里云容器服务和容器网络文件系统搭建WordPress网站
    • 下一篇
    阿里云对象存储OSS怎么收费？包年包月和按量付费价格表