Caffes 21 天-在工程中使用caffe库

VS2015

新建VS工程，根据自己实际的目录进行下面的配置(即类似于Caffe工程中matlab子工程的配置拷贝）。

(1)配置包含目录：

E:\caffe-windows\src
E:\caffe-windows\scripts\build\libraries\include\boost-1_61
E:\caffe-windows\scripts\build\libraries\include
E:\caffe-windows\scripts\build\include
E:\caffe-windows\scripts\build\libraries\include\opencv
D:\python\python2.7.12\include
D:\python\python2.7.12\lib\site-packages\numpy\core\include
E:\caffe-windows\include
E:\caffe-windows\scripts\build
E:\caffe-windows\scripts\build\libraries\Include

(2)Release 配置附加依赖项

..\build\lib\Release\caffe.lib
..\build\lib\Release\proto.lib
..\build\libraries\lib\boost_system-vc140-mt-1_61.lib
..\build\libraries\lib\boost_thread-vc140-mt-1_61.lib
..\build\libraries\lib\boost_filesystem-vc140-mt-1_61.lib
..\build\libraries\lib\boost_chrono-vc140-mt-1_61.lib
..\build\libraries\lib\boost_date_time-vc140-mt-1_61.lib
..\build\libraries\lib\boost_atomic-vc140-mt-1_61.lib
..\build\libraries\lib\boost_python-vc140-mt-1_61.lib
..\build\libraries\lib\glog.lib
..\build\libraries\lib\gflags.lib
shlwapi.lib
..\build\libraries\lib\libprotobuf.lib
..\build\libraries\lib\caffehdf5_hl.lib
..\build\libraries\lib\caffehdf5.lib
..\build\libraries\lib\lmdb.lib
ntdll.lib
..\build\libraries\lib\leveldb.lib
..\build\libraries\lib\snappy_static.lib
..\build\libraries\lib\caffezlib.lib
..\build\libraries\x64\vc14\lib\opencv_highgui310.lib
..\build\libraries\x64\vc14\lib\opencv_imgcodecs310.lib
..\build\libraries\x64\vc14\lib\opencv_imgproc310.lib
..\build\libraries\x64\vc14\lib\opencv_core310.lib
..\build\libraries\lib\libopenblas.dll.a
D:\python\python2.7.12\libs\python27.lib
..\build\libraries\lib\libboost_filesystem-vc140-mt-1_61.lib
..\build\libraries\lib\libboost_system-vc140-mt-1_61.lib
..\build\libraries\lib\libboost_date_time-vc140-mt-1_61.lib

(3)Debug 附加依赖项

..\build\lib\Debug\caffe-d.lib
..\build\lib\Debug\proto-d.lib

..\build\libraries\lib\libboost_system-vc140-mt-gd-1_61.lib
..\build\libraries\lib\boost_thread-vc140-mt-gd-1_61.lib
..\build\libraries\lib\libboost_filesystem-vc140-mt-gd-1_61.lib
..\build\libraries\lib\boost_chrono-vc140-mt-gd-1_61.lib
..\build\libraries\lib\libboost_date_time-vc140-mt-gd-1_61.lib
..\build\libraries\lib\boost_atomic-vc140-mt-gd-1_61.lib
..\build\libraries\lib\boost_python-vc140-mt-gd-1_61.lib

..\build\libraries\lib\glogd.lib
..\build\libraries\lib\gflagsd.lib

..\build\libraries\lib\libprotobufd.lib
..\build\libraries\lib\caffehdf5_hl_D.lib
..\build\libraries\lib\caffehdf5_D.lib
..\build\libraries\lib\lmdbd.lib

..\build\libraries\lib\leveldbd.lib

..\build\libraries\lib\snappy_staticd.lib
..\build\libraries\lib\caffezlibd.lib
..\build\libraries\x64\vc14\lib\opencv_core310d.lib
..\build\libraries\x64\vc14\lib\opencv_imgproc310d.lib
..\build\libraries\x64\vc14\lib\opencv_imgcodecs310d.lib
..\build\libraries\x64\vc14\lib\opencv_highgui310d.lib

..\build\libraries\lib\libopenblas.dll.a
D:\python\python2.7.12\libs\python27.lib

shlwapi.lib
ntdll.lib

设置Debug模式，我们就可以调试Caffe代码，跟踪到源程序，来深入的学习Caffe.

(4) 环境变量(dll)

E:\caffe-windows\scripts\build\tools\Release
E:\caffe-windows\scripts\build\libraries\bin
E:\caffe-windows\scripts\build\libraries\lib
E:\caffe-windows\scripts\build\libraries\x64\vc14\bin

1. get_param_from_proto.cpp

// VS2015 
#include "stdafx.h"

#define CPU_ONLY

/* caffe.ph.h 和caffe.pb.cc用于解析Caffe参数配置文件，将模型权值序列化和反序列化到磁盘的协议接口*/
/* add :  _CRT_SECURE_NO_WARNINGS 到预处理命令 */

#include
#include
#include
#include

#include
#include
using namespace std;
using google::protobuf::io::FileInputStream;
using google::protobuf::io::FileOutputStream;
using google::protobuf::io::ZeroCopyInputStream;
using google::protobuf::io::CodedInputStream;
using google::protobuf::io::ZeroCopyOutputStream;
using google::protobuf::io::CodedOutputStream;
using google::protobuf::Message;

#include
#include //for windows
int main(void)
{
    const char* filename = "solver.prototxt";
    caffe::SolverParameter solver_param;
    FILE * fp = fopen(filename, "r");

    int fd = _open(filename, O_RDONLY);
    if(fp == NULL)
    {
        cout << "File not found :" << filename << endl;

    }
    FileInputStream * input = new FileInputStream(fd);
    bool success = google::protobuf::TextFormat::Parse(input, &solver_param);
    delete input;
    _close(fd);
    cout << "Solver Mode = " << solver_param.solver_mode() << endl;
    cout << "Device id = " << solver_param.device_id() << endl;
    cout << "Solver Type = " << solver_param.solver_type() << endl;
    cout << "Rondom Seed = " << solver_param.random_seed() << endl;
    //cout << "Train net param = " << solver_param.train_net_param() << endl; 
    cout << "Max iter = " << solver_param.max_iter() << endl;
    cout << "Test interval = " << solver_param.test_interval() << endl;
    cout << "Momentum = " << solver_param.momentum()<< endl;
    cout << "gamam = " << solver_param.gamma() << endl;
    cout << "End " << endl;
}

result:

blob_demo.cpp

#include "stdafx.h"

#define CPU_ONLY //没有安装cuda
#include
#include
#include

/* Tip: libboost_filesystem-vc140-mt-1_61.lib,将其添加*/
#include //write blob to proto

using namespace caffe;
using namespace std;

int main(void)
{
    Blob<float> a;
    cout << "Size : " << a.shape_string() << endl;

    a.Reshape(1, 2, 3, 4);
    cout << "Size: " << a.shape_string() << endl;

    //change value of a 
    float *p = a.mutable_cpu_data();
    for (int i = 0; i < a.count(); i++)
    {
        p[i] = i;
    }
    cout << "a.count = " << a.count() << endl;
    //print a
    for (int u = 0; u < a.num(); u++) // image number 
    {
        for (int v = 0; v < a.channels(); v++) //image channels
        {
            for (int y = 0; y < a.height(); y++) // image height
            {
                for (int x = 0; x < a.width(); x++) // image width
                {
                    cout << "a[" << u << "][" << v << "][" << y << "][" << x << "] = " << a.data_at(u, v, y, x) << endl;
                }
            }
        }
    }
    // L1 norm and L2 norm
    cout << "ASUM = " << a.asum_data() << endl; //abs sum
    cout << "SUMSQ = " << a.sumsq_data() << endl; // sum sqrt

    /* change blob's diff*/
    float *q = a.mutable_cpu_diff();
    for (int i = 0; i < a.count(); i++)
    {
        q[i] = a.count() - 1 - i;
    }
    a.Update(); // a = a - diff
    //print a
    for (int u = 0; u < a.num(); u++) // image number 
    {
        for (int v = 0; v < a.channels(); v++) //image channels
        {
            for (int y = 0; y < a.height(); y++) // image height
            {
                for (int x = 0; x < a.width(); x++) // image width
                {
                    cout << "a[" << u << "][" << v << "][" << y << "][" << x << "] = " << a.data_at(u, v, y, x) << endl;
                }
            }
        }
    }
    /*将Blob内部的值保存到磁盘或者从磁盘载入内存,可以分别通过ToProto(),FromProto()实现*/
    BlobProto bp;
    a.ToProto(&bp, true);// 将a序列化，连同diff（默认不带)
    WriteProtoToBinaryFile(bp, "a.blob");
    BlobProto bp2;
    ReadProtoFromBinaryFileOrDie("a.blob", &bp2);
    Blob<float> b;
    b.FromProto(bp2, true);
    // print b

    for (int u = 0; u < a.num(); u++) // image number 
    {
        for (int v = 0; v < a.channels(); v++) //image channels
        {
            for (int y = 0; y < a.height(); y++) // image height
            {
                for (int x = 0; x < a.width(); x++) // image width
                {
                    cout << "b[" << u << "][" << v << "][" << y << "][" << x << "] = " << b.data_at(u, v, y, x) << endl;
                }
            }
        }
    }
    return 0;
}

caffe_reg.h

#pragma once
#ifndef CAFFE_REG_H
#define CAFFE_REG_H

#include "caffe/common.hpp"  
#include "caffe/layers/input_layer.hpp"  
#include "caffe/layers/inner_product_layer.hpp"  
#include "caffe/layers/dropout_layer.hpp"  
#include "caffe/layers/conv_layer.hpp"  
#include "caffe/layers/relu_layer.hpp"  

#include "caffe/layers/pooling_layer.hpp"  
#include "caffe/layers/lrn_layer.hpp"  
#include "caffe/layers/softmax_layer.hpp"  
//#include "caffe/layers/normalize_layer.hpp"
//#include "caffe/layers/permute_layer.hpp"
#include "caffe/layers/flatten_layer.hpp"
//#include "caffe/layers/prior_box_layer.hpp"
#include "caffe/layers/concat_layer.hpp"
#include "caffe/layers/reshape_layer.hpp"
#include "caffe/layers/softmax_layer.hpp"
//#include "caffe/layers/detection_output_layer.hpp"

namespace caffe
{
    extern INSTANTIATE_CLASS(InputLayer);
    //REGISTER_LAYER_CLASS(Input);
    //extern INSTANTIATE_CLASS(ConvolutionLayer);
    //REGISTER_LAYER_CLASS(Convolution);
    extern INSTANTIATE_CLASS(InnerProductLayer);
    extern INSTANTIATE_CLASS(DropoutLayer);

    //extern INSTANTIATE_CLASS(ReLULayer);
    //REGISTER_LAYER_CLASS(ReLU);
    //extern INSTANTIATE_CLASS(PoolingLayer);
    //REGISTER_LAYER_CLASS(Pooling);
    //extern INSTANTIATE_CLASS(LRNLayer);
    //REGISTER_LAYER_CLASS(LRN);
    //extern INSTANTIATE_CLASS(SoftmaxLayer);
    //REGISTER_LAYER_CLASS(Softmax);
    //extern INSTANTIATE_CLASS(NormalizeLayer);
    //REGISTER_LAYER_CLASS(Normalize);
    //extern INSTANTIATE_CLASS(PermuteLayer);
    //REGISTER_LAYER_CLASS(Permute);
    //extern INSTANTIATE_CLASS(FlattenLayer);
    //extern INSTANTIATE_CLASS(PriorBoxLayer);
    //extern INSTANTIATE_CLASS(ConcatLayer);
    //extern INSTANTIATE_CLASS(ReshapeLayer);
    //extern INSTANTIATE_CLASS(SoftmaxLayer);
    //REGISTER_LAYER_CLASS(Softmax);
    //extern INSTANTIATE_CLASS(DetectionOutputLayer);
}

#endif
//http://blog.csdn.net/birdwcp/article/details/53580068

net_demo.cpp

#include"stdafx.h"
#define  CPU_ONLY
#include"caffe_reg.h"
#include
#include
#include
using namespace caffe;
using namespace std;

int main(void)
{
    string proto("deploy.prototxt");
    Net<float> nn(proto, caffe::TEST);
    vector<string> bn = nn.blob_names(); //获得Net中所有Blob对象
    vector<string> ln = nn.layer_names();
    for (int i = 0; i < bn.size(); i++)
    {
        cout << "Blob # " << i << ":" << bn[i] << endl;
    }

    for (int i = 0; i < ln.size(); i++)
    {
        cout << "Layer # " << i << ":" << ln[i] << endl;
    }
    return 0; 

}

从上面的例子我们可以看出，Net中即包括了Layer对象，又包括了Blob对象。其中Blob对象用于存放每个Layer输入/输出的中间结果，Layer则根据Net描述对指定的输入Blob进行某种计算处理（卷积,下采样,全连接,非线性变换,计算代价函数等）,输出的结果放在指定的Blob中，由于输入Blob和输出Blob可能为同一个，所以上面给出的Layer24个，而Blob只有15个。

补充：

caffe.lib 静态库的编译: 仅需要头文件和源文件。
C++/常规/附加包含目录。

F:\vs\caffe-windows\src;
F:\vs\caffe-windows\scripts\build\libraries\include\boost-1_61;
F:\vs\caffe-windows\scripts\build\libraries\include;
F:\vs\caffe-windows\scripts\build\include;
C:\Program Files\NVIDIA GPU Computing Toolkit\CUDA\v8.0\include;
F:\vs\caffe-windows\scripts\build\libraries\include\opencv;
D:\Python27\include;
D:\Python27\lib\site-packages\numpy\core\include;
F:\vs\caffe-windows\include;
F:\vs\caffe-windows\scripts\build;
F:\vs\caffe-windows\scripts\build\libraries\Include;
%(AdditionalIncludeDirectories)

编译动态库时.pyd，如：会把连接的静态库编译进去，因此需要 【连接器-输入-附加依赖项】。

参考文献
1.深度学习21天实战Caffe.pdf