Caffe SourceCode Learning-main

Caffe的windows安装和Linux安装在这里就不介绍了，网上有很多很好的教程。

 

1.Caffe框架的入口为./tools/caffe.cpp的main()

（1）框架使用gflag对命令行参数进行解析

DEFINE_type(p1,p2,p3)定义了需要从命令行传递参数的变量：p1为变量，p2默认值，p3为解释。GlobalInit()函数中调用了ParseCommandLineFlags()，该函数是为了解析命令行参数，执行完了该函数，就可以在代码中使用FLAGS_p1使用该参数了，SetUsageMessage函数用于设置命令行帮助信息（-help）。

DEFINE_string(gpu, "",
    "Optional; run in GPU mode on given device IDs separated by ','."
    "Use '-gpu all' to run on all available GPUs. The effective training "
    "batch size is multiplied by the number of devices.");
DEFINE_string(solver, "",
    "The solver definition protocol buffer text file.");
DEFINE_string(model, "",
    "The model definition protocol buffer text file.");
DEFINE_string(phase, "",
    "Optional; network phase (TRAIN or TEST). Only used for 'time'.");
DEFINE_int32(level, 0,
    "Optional; network level.");
DEFINE_string(stage, "",
    "Optional; network stages (not to be confused with phase), "
    "separated by ','.");
DEFINE_string(snapshot, "",
    "Optional; the snapshot solver state to resume training.");
DEFINE_string(weights, "",
    "Optional; the pretrained weights to initialize finetuning, "
    "separated by ','. Cannot be set simultaneously with snapshot.");
DEFINE_int32(iterations, 50,
    "The number of iterations to run.");
DEFINE_string(sigint_effect, "stop",
             "Optional; action to take when a SIGINT signal is received: "
              "snapshot, stop or none.");
DEFINE_string(sighup_effect, "snapshot",
             "Optional; action to take when a SIGHUP signal is received: "
             "snapshot, stop or none.");

// Set version
  gflags::SetVersionString(AS_STRING(CAFFE_VERSION));
  // Usage message.
  gflags::SetUsageMessage("command line brew\n"
      "usage: caffe <command> <args>\n\n"
      "commands:\n"
      "  train           train or finetune a model\n"
      "  test            score a model\n"
      "  device_query    show GPU diagnostic information\n"
      "  time            benchmark model execution time");
  // Run tool or show usage.
  caffe::GlobalInit(&argc, &argv);

（2）各执行函数的注册

在train() test() time() device_query()四个函数定义的后面，各自通过注册器RegisterBrewFunction("xxx") 来注册函数到容器里。（RegisterBrewFunction(train)）

// A simple registry for caffe commands.
typedef int (*BrewFunction)();  // 定义函数指针类型 （指向参数为空、返回值为int的函数）
typedef std::map<caffe::string, BrewFunction> BrewMap; 
BrewMap g_brew_map;

#define RegisterBrewFunction(func) \             #注册器
namespace { \
class __Registerer_##func { \
 public: /* NOLINT */ \
  __Registerer_##func() { \
    g_brew_map[#func] = &func; \                  #将每一个<函数名，函数指针>注入g_brew_map
  } \
}; \
__Registerer_##func g_registerer_##func; \
}

argv[1]) = "train"

return GetBrewFunction(caffe::string(argv[1]))();  //return(train())

2.进入具体的执行函数train()

（1）根据FLAGS_solver读取超参信息

caffe::SolverParameter solver_param;
caffe::ReadSolverParamsFromTextFileOrDie(FLAGS_solver, &solver_param);  

-->>ReadSolverParamsFromTextFileOrDie()    #upgrade_proto.cpp

    -->> bool ReadProtoFromTextFile(const string& filename, Message* proto)   #io.cpp

（2）通过SolverFactory创建一个Solver类

shared_ptr<caffe::Solver<float> >  solver(caffe::SolverRegistry<float>::CreateSolver(solver_param));

（3）执行迭代

solver->Solve();