CNN c++代码（tiny_cnn）学习

来源：互联网发布：centos加固编辑：程序博客网时间：2024/05/21 17:28

注：MNIST数据集格式 http://blog.csdn.net/liulina603/article/details/48434947

1. 连载的分析tiny_cnn的文章 http://blog.csdn.net/u013088062/article/details/50839015 配置不了（头文件包含不成功）

2. http://blog.csdn.net/hao529good/article/details/45419031

结果图如下：要跑5分钟（60000个样本训练2遍、10000个样本测试2遍）

main.cpp如下

#include "iostream"#include "util.h"#include "vector"#include "CNN.h"#include "time.h"#include "stdlib.h"using namespace std;// 把MNIST的四个数据集放在根目录下，放在别处ifstream可能会打不开文件int main(){// initialize input data double*** train_x, ***test_x;double **train_label, **test_label;int NumberOfImages = 60000;int length = 28;train_x = new double**[NumberOfImages];test_x = new double**[NumberOfImages/6];train_label = new double *[NumberOfImages];test_label = new double *[NumberOfImages/6];for (int i=0; i<NumberOfImages; i++){train_x[i]=new double*[28];train_label[i]=new double[10];if (i<NumberOfImages/6){test_x[i]=new double*[28];test_label[i] = new double [10];}for (int j=0; j<28; j++){train_x[i][j]=new double[28];if (i<NumberOfImages/6){test_x[i][j]=new double[28];}}} ReadMNIST(NumberOfImages,length,train_x,"train-images.idx3-ubyte"); ReadMNIST_Label(NumberOfImages,train_label, "train-labels.idx1-ubyte");ReadMNIST(NumberOfImages/6,length,test_x,"t10k-images.idx3-ubyte");ReadMNIST_Label(NumberOfImages/6,test_label, "t10k-labels.idx1-ubyte");    // constructor CNNLayerBuilder builder;    Layer layer;    builder.addLayer(layer.buildInputLayer(size(28,28)));    builder.addLayer(layer.buildConvLayer(6, size(5, 5)));    builder.addLayer(layer.buildSampLayer( size(2, 2)));    builder.addLayer(layer.buildConvLayer(12, size(5, 5)));    builder.addLayer(layer.buildSampLayer( size(2, 2)));    //builder.addLayer(layer.buildConvLayer(20, size(4, 4)));    builder.addLayer(layer.buildOutputLayer(10));    CNN cnn = CNN(builder, 10);// biuder batchsize    for (int i=0; i<2; i++)    {double t1 = cpu_time();cnn.train(train_x,train_label, NumberOfImages);double t2 = cpu_time();cout<<t2-t1<<" s"<<endl;cnn.test(test_x,test_label, NumberOfImages/6);    }// delete datafor (int i=0; i<NumberOfImages; i++){delete []train_label[i];for (int j=0; j<28; j++){delete []train_x[i][j];}delete []train_x[i];}for (int i=0; i<NumberOfImages/6; i++){delete []test_label[i];for (int j=0; j<28; j++){delete []test_x[i][j];}delete []test_x[i];}delete []train_label;delete []train_x;delete []test_x;delete []test_label;return 0;}注释之后再重新发上来。

0 0