[置顶] caffe源码解析 — caffe.proto

引言

要看caffe源码，我认为首先应该看的就是caffe.proto。
它位于…\src\caffe\proto目录下，在这个文件夹下还有一个.pb.cc和一个.pb.h文件，这两个文件都是由caffe.proto编译而来的。
在caffe.proto中定义了很多结构化数据，包括：

• BlobProto
• Datum
• FillerParameter
• NetParameter
• SolverParameter
• SolverState
• LayerParameter
• ConcatParameter
• ConvolutionParameter
• DataParameter
• DropoutParameter
• HDF5DataParameter
• HDF5OutputParameter
• ImageDataParameter
• InfogainLossParameter
• InnerProductParameter
• LRNParameter
• MemoryDataParameter
• PoolingParameter
• PowerParameter
• WindowDataParameter
• V0LayerParameter

正文

一、什么是protocol buffer

以下内容摘自：Google Protocol Buffer 的使用和原理
强烈推荐另外一篇极好的博文是：Protocol Buffer技术详解(C++实例)

简介

什么是 Google Protocol Buffer？ 假如您在网上搜索，应该会得到类似这样的文字介绍：
Google Protocol Buffer( 简称 Protobuf) 是 Google 公司内部的混合语言数据标准，目前已经正在使用的有超过 48,162 种报文格式定义和超过 12,183 个 .proto 文件。他们用于 RPC 系统和持续数据存储系统。
Protocol Buffers 是一种轻便高效的结构化数据存储格式，可以用于结构化数据串行化，或者说序列化。它很适合做数据存储或 RPC 数据交换格式。可用于通讯协议、数据存储等领域的语言无关、平台无关、可扩展的序列化结构数据格式。目前提供了 C++、Java、Python 三种语言的 API。
或许您和我一样，在第一次看完这些介绍后还是不明白 Protobuf 究竟是什么，那么我想一个简单的例子应该比较有助于理解它。

一个简单的例子

安装 Google Protocol Buffer
在网站 http://code.google.com/p/protobuf/downloads/list上可以下载 Protobuf 的源代码。然后解压编译安装便可以使用它了。
安装步骤如下所示：

 tar -xzf protobuf-2.1.0.tar.gz 
 cd protobuf-2.1.0 
 ./configure --prefix=$INSTALL_DIR 
 make 
 make check 
 make install

关于简单例子的描述

我打算使用 Protobuf 和 C++ 开发一个十分简单的例子程序。
该程序由两部分组成。第一部分被称为 Writer，第二部分叫做 Reader。
Writer 负责将一些结构化的数据写入一个磁盘文件，Reader 则负责从该磁盘文件中读取结构化数据并打印到屏幕上。
准备用于演示的结构化数据是 HelloWorld，它包含两个基本数据：

• ID，为一个整数类型的数据
• Str，这是一个字符串

书写 .proto 文件

首先我们需要编写一个 proto 文件，定义我们程序中需要处理的结构化数据，在 protobuf 的术语中，结构化数据被称为 Message。proto 文件非常类似 java 或者 C 语言的数据定义。代码清单 1 显示了例子应用中的 proto 文件内容。
清单 1. proto 文件

package lm; 
 message helloworld 
 { 
    required int32     id = 1;  // ID 
    required string    str = 2;  // str 
    optional int32     opt = 3;  //optional field 
 }

一个比较好的习惯是认真对待 proto 文件的文件名。比如将命名规则定于
packageName.MessageName.proto
在上例中，package 名字叫做 lm，定义了一个消息 helloworld，该消息有三个成员，类型为 int32 的 id，另一个为类型为 string 的成员 str。opt 是一个可选的成员，即消息中可以不包含该成员。

编译 .proto 文件

写好 proto 文件之后就可以用 Protobuf 编译器将该文件编译成目标语言了。本例中我们将使用 C++。
假设您的 proto 文件存放在 $SRC_DIR 下面，您也想把生成的文件放在同一个目录下，则可以使用如下命令：

protoc -I=$SRC_DIR --cpp_out=$DST_DIR $SRC_DIR/addressbook.proto

命令将生成两个文件：
lm.helloworld.pb.h ， 定义了 C++ 类的头文件
lm.helloworld.pb.cc ， C++ 类的实现文件
在生成的头文件中，定义了一个 C++ 类 helloworld，后面的 Writer 和 Reader 将使用这个类来对消息进行操作。诸如对消息的成员进行赋值，将消息序列化等等都有相应的方法。

编写 writer 和 Reader

如前所述，Writer将把一个结构化数据写入磁盘，以便其他人来读取。假如我们不使用 Protobuf，其实也有许多的选择。一个可能的方法是将数据转换为字符串，然后将字符串写入磁盘。转换为字符串的方法可以使用sprintf()，这非常简单。数字123可以变成字符串“123”。
这样做似乎没有什么不妥，但是仔细考虑一下就会发现，这样的做法对写 Reader 的那个人的要求比较高，Reader 的作者必须了 Writer 的细节。比如”123”可以是单个数字 123，但也可以是三个数字 1,2 和 3，等等。这么说来，我们还必须让 Writer 定义一种分隔符一样的字符，以便 Reader 可以正确读取。但分隔符也许还会引起其他的什么问题。最后我们发现一个简单的 Helloworld 也需要写许多处理消息格式的代码。
如果使用 Protobuf，那么这些细节就可以不需要应用程序来考虑了。
使用 Protobuf，Writer 的工作很简单，需要处理的结构化数据由 .proto 文件描述，经过上一节中的编译过程后，该数据化结构对应了一个 C++ 的类，并定义在 lm.helloworld.pb.h 中。对于本例，类名为 lm::helloworld。
Writer 需要 include 该头文件，然后便可以使用这个类了。
现在，在 Writer 代码中，将要存入磁盘的结构化数据由一个 lm::helloworld 类的对象表示，它提供了一系列的 get/set 函数用来修改和读取结构化数据中的数据成员，或者叫 field。
当我们需要将该结构化数据保存到磁盘上时，类 lm::helloworld 已经提供相应的方法来把一个复杂的数据变成一个字节序列，我们可以将这个字节序列写入磁盘。
对于想要读取这个数据的程序来说，也只需要使用类 lm::helloworld 的相应反序列化方法来将这个字节序列重新转换会结构化数据。这同我们开始时那个“123”的想法类似，不过 Protobuf 想的远远比我们那个粗糙的字符串转换要全面，因此，我们不如放心将这类事情交给 Protobuf 吧。
程序清单 2 演示了 Writer 的主要代码，您一定会觉得很简单吧？
清单 2. Writer 的主要代码

 #include "lm.helloworld.pb.h"
…

 int main(void) 
 { 

  lm::helloworld msg1; 
  msg1.set_id(101); 
  msg1.set_str(“hello”); 

  // Write the new address book back to disk. 
  fstream output("./log", ios::out | ios::trunc | ios::binary); 

  if (!msg1.SerializeToOstream(&output)) { 
      cerr << "Failed to write msg." << endl; 
      return -1; 
  }         
  return 0; 
 }

Msg1 是一个 helloworld 类的对象，set_id() 用来设置 id 的值。SerializeToOstream 将对象序列化后写入一个 fstream 流。
代码清单 3 列出了 reader 的主要代码。
清单 3. Reader

 #include "lm.helloworld.pb.h" 
…
 void ListMsg(const lm::helloworld & msg) { 
  cout << msg.id() << endl; 
  cout << msg.str() << endl; 
 } 

 int main(int argc, char* argv[]) { 

  lm::helloworld msg1; 

  { 
    fstream input("./log", ios::in | ios::binary); 
    if (!msg1.ParseFromIstream(&input)) { 
      cerr << "Failed to parse address book." << endl; 
      return -1; 
    } 
  } 

  ListMsg(msg1); 
  … 
 }

同样，Reader 声明类 helloworld 的对象 msg1，然后利用 ParseFromIstream 从一个 fstream 流中读取信息并反序列化。此后，ListMsg 中采用 get 方法读取消息的内部信息，并进行打印输出操作。
运行结果
运行 Writer 和 Reader 的结果如下：

\>writer 
\>reader 
101 
Hello

Reader 读取文件 log 中的序列化信息并打印到屏幕上。本文中所有的例子代码都可以在附件中下载。您可以亲身体验一下。
这个例子本身并无意义，但只要您稍加修改就可以将它变成更加有用的程序。比如将磁盘替换为网络 socket，那么就可以实现基于网络的数据交换任务。而存储和交换正是 Protobuf 最有效的应用领域。

二、caffe.proto中的几个重要数据类型

看完了上面关于protocol buffer的介绍，大家应该可以知道其实caffe.pb.cc里面的东西都是从caffe.proto编译而来的，无非就是一些关于这些数据结构（类）的标准化操作，比如

  void CopyFrom();
  void MergeFrom();
  void CopyFrom();
  void MergeFrom;
  void Clear();
  bool IsInitialized() const;
  int ByteSize() const;
  bool MergePartialFromCodedStream();
  void SerializeWithCachedSizes() const;
  SerializeWithCachedSizesToArray() const;
  int GetCachedSize()
  void SharedCtor();
  void SharedDtor();
  void SetCachedSize() const;

<0> BlobProto

message BlobProto {//blob的属性以及blob中的数据(data\diff)
  optional int32 num = 1 [default = 0];
  optional int32 channels = 2 [default = 0];
  optional int32 height = 3 [default = 0];
  optional int32 width = 4 [default = 0];
  repeated float data = 5 [packed = true];
  repeated float diff = 6 [packed = true];
}

<1> Datum

  message Datum {
  optional int32 channels = 1;
  optional int32 height = 2;
  optional int32 width = 3;
  optional bytes data = 4;//真实的图像数据，以字节存储(bytes)
  optional int32 label = 5;
  repeated float float_data = 6;//datum也能存float类型的数据(float)
}

<2> LayerParameter

message LayerParameter {
  repeated string bottom = 2; //输入的blob的名字(string)
  repeated string top = 3; //输出的blob的名字(string)
  optional string name = 4; //层的名字
  enum LayerType { //层的枚举（enum，和c++中的enum一样）
    NONE = 0;
    ACCURACY = 1;
    BNLL = 2;
    CONCAT = 3;
    CONVOLUTION = 4;
    DATA = 5;
    DROPOUT = 6;
    EUCLIDEAN_LOSS = 7;
    ELTWISE_PRODUCT = 25;
    FLATTEN = 8;
    HDF5_DATA = 9;
    HDF5_OUTPUT = 10;
    HINGE_LOSS = 28;
    IM2COL = 11;
    IMAGE_DATA = 12;
    INFOGAIN_LOSS = 13;
    INNER_PRODUCT = 14;
    LRN = 15;
    MEMORY_DATA = 29;
    MULTINOMIAL_LOGISTIC_LOSS = 16;
    POOLING = 17;
    POWER = 26;
    RELU = 18;
    SIGMOID = 19;
    SIGMOID_CROSS_ENTROPY_LOSS = 27;
    SOFTMAX = 20;
    SOFTMAX_LOSS = 21;
    SPLIT = 22;
    TANH = 23;
    WINDOW_DATA = 24;
  }
  optional LayerType type = 5; // 层的类型
  repeated BlobProto blobs = 6; //blobs的数值参数
  repeated float blobs_lr = 7; //学习速率(repeated)，如果你想那个设置一个blob的学习速率，你需要设置所有blob的学习速率。
  repeated float weight_decay = 8; //权值衰减(repeated)

  // 相对于某一特定层的参数(optional)
  optional ConcatParameter concat_param = 9;
  optional ConvolutionParameter convolution_param = 10;
  optional DataParameter data_param = 11;
  optional DropoutParameter dropout_param = 12;
  optional HDF5DataParameter hdf5_data_param = 13;
  optional HDF5OutputParameter hdf5_output_param = 14;
  optional ImageDataParameter image_data_param = 15;
  optional InfogainLossParameter infogain_loss_param = 16;
  optional InnerProductParameter inner_product_param = 17;
  optional LRNParameter lrn_param = 18;
  optional MemoryDataParameter memory_data_param = 22;
  optional PoolingParameter pooling_param = 19;
  optional PowerParameter power_param = 21;
  optional WindowDataParameter window_data_param = 20;
  optional V0LayerParameter layer = 1;
}

<3> NetParameter

message NetParameter {
  optional string name = 1;//网络的名字
  repeated LayerParameter layers = 2; //repeated类似于数组
  repeated string input = 3;//输入层blob的名字
  repeated int32 input_dim = 4;//输入层blob的维度，应该等于(4*#input)
  optional bool force_backward = 5 [default = false];//网络是否进行反向传播。如果设置为否，则由网络的结构和学习速率来决定是否进行反向传播。
}

<4> SolverParameter

message SolverParameter {
  optional string train_net = 1; // 训练网络的proto file
  optional string test_net = 2; // 测试网络的proto file
  optional int32 test_iter = 3 [default = 0]; // 每次测试时的迭代次数
  optional int32 test_interval = 4 [default = 0]; // 两次测试的间隔迭代次数
  optional bool test_compute_loss = 19 [default = false];
  optional float base_lr = 5; // 基本学习率
  optional int32 display = 6; // 两次显示的间隔迭代次数
  optional int32 max_iter = 7; // 最大迭代次数
  optional string lr_policy = 8; // 学习速率衰减方式
  optional float gamma = 9; // 关于梯度下降的一个参数
  optional float power = 10; // 计算学习率的一个参数
  optional float momentum = 11; // 动量
  optional float weight_decay = 12; // 权值衰减
  optional int32 stepsize = 13; // 学习速率的衰减步长
  optional int32 snapshot = 14 [default = 0]; // snapshot的间隔
  optional string snapshot_prefix = 15; // snapshot的前缀
  optional bool snapshot_diff = 16 [default = false]; // 是否对于 diff 进行 snapshot
  enum SolverMode {
    CPU = 0;
    GPU = 1;
  }
  optional SolverMode solver_mode = 17 [default = GPU]; // solver的模式，默认为GPU
  optional int32 device_id = 18 [default = 0]; // GPU的ID
  optional int64 random_seed = 20 [default = -1]; // 随机数种子
}

三、caffe.proto源码

// Copyright 2014 BVLC and contributors.
package caffe;

message BlobProto {
  optional int32 num = 1 [default = 0];
  optional int32 channels = 2 [default = 0];
  optional int32 height = 3 [default = 0];
  optional int32 width = 4 [default = 0];
  repeated float data = 5 [packed = true];
  repeated float diff = 6 [packed = true];
}

// The BlobProtoVector is simply a way to pass multiple blobproto instances
// around.
message BlobProtoVector {
  repeated BlobProto blobs = 1;
}

message Datum {
  optional int32 channels = 1;
  optional int32 height = 2;
  optional int32 width = 3;
  // the actual image data, in bytes
  optional bytes data = 4;
  optional int32 label = 5;
  // Optionally, the datum could also hold float data.
  repeated float float_data = 6;
}

message FillerParameter {
  // The filler type.
  optional string type = 1 [default = 'constant'];
  optional float value = 2 [default = 0]; // the value in constant filler
  optional float min = 3 [default = 0]; // the min value in uniform filler
  optional float max = 4 [default = 1]; // the max value in uniform filler
  optional float mean = 5 [default = 0]; // the mean value in Gaussian filler
  optional float std = 6 [default = 1]; // the std value in Gaussian filler
  // The expected number of non-zero input weights for a given output in
  // Gaussian filler -- the default -1 means don't perform sparsification.
  optional int32 sparse = 7 [default = -1];
}

message NetParameter {
  optional string name = 1; // consider giving the network a name
  repeated LayerParameter layers = 2; // a bunch of layers.
  // The input blobs to the network.
  repeated string input = 3;
  // The dim of the input blobs. For each input blob there should be four
  // values specifying the num, channels, height and width of the input blob.
  // Thus, there should be a total of (4 * #input) numbers.
  repeated int32 input_dim = 4;
  // Whether the network will force every layer to carry out backward operation.
  // If set False, then whether to carry out backward is determined
  // automatically according to the net structure and learning rates.
  optional bool force_backward = 5 [default = false];
}

message SolverParameter {
  optional string train_net = 1; // The proto file for the training net.
  optional string test_net = 2; // The proto file for the testing net.
  // The number of iterations for each testing phase.
  optional int32 test_iter = 3 [default = 0];
  // The number of iterations between two testing phases.
  optional int32 test_interval = 4 [default = 0];
  optional bool test_compute_loss = 19 [default = false];
  optional float base_lr = 5; // The base learning rate
  // the number of iterations between displaying info. If display = 0, no info
  // will be displayed.
  optional int32 display = 6;
  optional int32 max_iter = 7; // the maximum number of iterations
  optional string lr_policy = 8; // The learning rate decay policy.
  optional float gamma = 9; // The parameter to compute the learning rate.
  optional float power = 10; // The parameter to compute the learning rate.
  optional float momentum = 11; // The momentum value.
  optional float weight_decay = 12; // The weight decay.
  optional int32 stepsize = 13; // the stepsize for learning rate policy "step"
  optional int32 snapshot = 14 [default = 0]; // The snapshot interval
  optional string snapshot_prefix = 15; // The prefix for the snapshot.
  // whether to snapshot diff in the results or not. Snapshotting diff will help
  // debugging but the final protocol buffer size will be much larger.
  optional bool snapshot_diff = 16 [default = false];
  // the mode solver will use: 0 for CPU and 1 for GPU. Use GPU in default.
  enum SolverMode {
    CPU = 0;
    GPU = 1;
  }
  optional SolverMode solver_mode = 17 [default = GPU];
  // the device_id will that be used in GPU mode. Use device_id = 0 in default.
  optional int32 device_id = 18 [default = 0];
  // If non-negative, the seed with which the Solver will initialize the Caffe
  // random number generator -- useful for reproducible results. Otherwise,
  // (and by default) initialize using a seed derived from the system clock.
  optional int64 random_seed = 20 [default = -1];
}

// A message that stores the solver snapshots
message SolverState {
  optional int32 iter = 1; // The current iteration
  optional string learned_net = 2; // The file that stores the learned net.
  repeated BlobProto history = 3; // The history for sgd solvers
}

// Update the next available ID when you add a new LayerParameter field.
//
// LayerParameter next available ID: 23 (last added: memory_data_param)
message LayerParameter {
  repeated string bottom = 2; // the name of the bottom blobs
  repeated string top = 3; // the name of the top blobs
  optional string name = 4; // the layer name

  // Add new LayerTypes to the enum below in lexicographical order (other than
  // starting with NONE), starting with the next available ID in the comment
  // line above the enum. Update the next available ID when you add a new
  // LayerType.
  //
  // LayerType next available ID: 30 (last added: MEMORY_DATA)
  enum LayerType {
    // "NONE" layer type is 0th enum element so that we don't cause confusion
    // by defaulting to an existent LayerType (instead, should usually error if
    // the type is unspecified).
    NONE = 0;
    ACCURACY = 1;
    BNLL = 2;
    CONCAT = 3;
    CONVOLUTION = 4;
    DATA = 5;
    DROPOUT = 6;
    EUCLIDEAN_LOSS = 7;
    ELTWISE_PRODUCT = 25;
    FLATTEN = 8;
    HDF5_DATA = 9;
    HDF5_OUTPUT = 10;
    HINGE_LOSS = 28;
    IM2COL = 11;
    IMAGE_DATA = 12;
    INFOGAIN_LOSS = 13;
    INNER_PRODUCT = 14;
    LRN = 15;
    MEMORY_DATA = 29;
    MULTINOMIAL_LOGISTIC_LOSS = 16;
    POOLING = 17;
    POWER = 26;
    RELU = 18;
    SIGMOID = 19;
    SIGMOID_CROSS_ENTROPY_LOSS = 27;
    SOFTMAX = 20;
    SOFTMAX_LOSS = 21;
    SPLIT = 22;
    TANH = 23;
    WINDOW_DATA = 24;
  }
  optional LayerType type = 5; // the layer type from the enum above

  // The blobs containing the numeric parameters of the layer
  repeated BlobProto blobs = 6;
  // The ratio that is multiplied on the global learning rate. If you want to
  // set the learning ratio for one blob, you need to set it for all blobs.
  repeated float blobs_lr = 7;
  // The weight decay that is multiplied on the global weight decay.
  repeated float weight_decay = 8;

  // Parameters for particular layer types.
  optional ConcatParameter concat_param = 9;
  optional ConvolutionParameter convolution_param = 10;
  optional DataParameter data_param = 11;
  optional DropoutParameter dropout_param = 12;
  optional HDF5DataParameter hdf5_data_param = 13;
  optional HDF5OutputParameter hdf5_output_param = 14;
  optional ImageDataParameter image_data_param = 15;
  optional InfogainLossParameter infogain_loss_param = 16;
  optional InnerProductParameter inner_product_param = 17;
  optional LRNParameter lrn_param = 18;
  optional MemoryDataParameter memory_data_param = 22;
  optional PoolingParameter pooling_param = 19;
  optional PowerParameter power_param = 21;
  optional WindowDataParameter window_data_param = 20;

  // DEPRECATED: The layer parameters specified as a V0LayerParameter.
  // This should never be used by any code except to upgrade to the new
  // LayerParameter specification.
  optional V0LayerParameter layer = 1;
}

// Message that stores parameters used by ConcatLayer
message ConcatParameter {
  // Concat Layer needs to specify the dimension along the concat will happen,
  // the other dimensions must be the same for all the bottom blobs
  // By default it will concatenate blobs along channels dimension
  optional uint32 concat_dim = 1 [default = 1];
}

// Message that stores parameters used by ConvolutionLayer
message ConvolutionParameter {
  optional uint32 num_output = 1; // The number of outputs for the layer
  optional bool bias_term = 2 [default = true]; // whether to have bias terms
  optional uint32 pad = 3 [default = 0]; // The padding size
  optional uint32 kernel_size = 4; // The kernel size
  optional uint32 group = 5 [default = 1]; // The group size for group conv
  optional uint32 stride = 6 [default = 1]; // The stride
  optional FillerParameter weight_filler = 7; // The filler for the weight
  optional FillerParameter bias_filler = 8; // The filler for the bias
}

// Message that stores parameters used by DataLayer
message DataParameter {
  // Specify the data source.
  optional string source = 1;
  // For data pre-processing, we can do simple scaling and subtracting the
  // data mean, if provided. Note that the mean subtraction is always carried
  // out before scaling.
  optional float scale = 2 [default = 1];
  optional string mean_file = 3;
  // Specify the batch size.
  optional uint32 batch_size = 4;
  // Specify if we would like to randomly crop an image.
  optional uint32 crop_size = 5 [default = 0];
  // Specify if we want to randomly mirror data.
  optional bool mirror = 6 [default = false];
  // The rand_skip variable is for the data layer to skip a few data points
  // to avoid all asynchronous sgd clients to start at the same point. The skip
  // point would be set as rand_skip * rand(0,1). Note that rand_skip should not
  // be larger than the number of keys in the leveldb.
  optional uint32 rand_skip = 7 [default = 0];
}

// Message that stores parameters used by DropoutLayer
message DropoutParameter {
  optional float dropout_ratio = 1 [default = 0.5]; // dropout ratio
}

// Message that stores parameters used by HDF5DataLayer
message HDF5DataParameter {
  // Specify the data source.
  optional string source = 1;
  // Specify the batch size.
  optional uint32 batch_size = 2;
}

// Message that stores parameters used by HDF5OutputLayer
message HDF5OutputParameter {
  optional string file_name = 1;
}

// Message that stores parameters used by ImageDataLayer
message ImageDataParameter {
  // Specify the data source.
  optional string source = 1;
  // For data pre-processing, we can do simple scaling and subtracting the
  // data mean, if provided. Note that the mean subtraction is always carried
  // out before scaling.
  optional float scale = 2 [default = 1];
  optional string mean_file = 3;
  // Specify the batch size.
  optional uint32 batch_size = 4;
  // Specify if we would like to randomly crop an image.
  optional uint32 crop_size = 5 [default = 0];
  // Specify if we want to randomly mirror data.
  optional bool mirror = 6 [default = false];
  // The rand_skip variable is for the data layer to skip a few data points
  // to avoid all asynchronous sgd clients to start at the same point. The skip
  // point would be set as rand_skip * rand(0,1). Note that rand_skip should not
  // be larger than the number of keys in the leveldb.
  optional uint32 rand_skip = 7 [default = 0];
  // Whether or not ImageLayer should shuffle the list of files at every epoch.
  optional bool shuffle = 8 [default = false];
  // It will also resize images if new_height or new_width are not zero.
  optional uint32 new_height = 9 [default = 0];
  optional uint32 new_width = 10 [default = 0];
}

// Message that stores parameters InfogainLossLayer
message InfogainLossParameter {
  // Specify the infogain matrix source.
  optional string source = 1;
}

// Message that stores parameters used by InnerProductLayer
message InnerProductParameter {
  optional uint32 num_output = 1; // The number of outputs for the layer
  optional bool bias_term = 2 [default = true]; // whether to have bias terms
  optional FillerParameter weight_filler = 3; // The filler for the weight
  optional FillerParameter bias_filler = 4; // The filler for the bias
}

// Message that stores parameters used by LRNLayer
message LRNParameter {
  optional uint32 local_size = 1 [default = 5];
  optional float alpha = 2 [default = 1.];
  optional float beta = 3 [default = 0.75];
  enum NormRegion {
    ACROSS_CHANNELS = 0;
    WITHIN_CHANNEL = 1;
  }
  optional NormRegion norm_region = 4 [default = ACROSS_CHANNELS];
}

// Message that stores parameters used by MemoryDataLayer
message MemoryDataParameter {
  optional uint32 batch_size = 1;
  optional uint32 channels = 2;
  optional uint32 height = 3;
  optional uint32 width = 4;
}

// Message that stores parameters used by PoolingLayer
message PoolingParameter {
  enum PoolMethod {
    MAX = 0;
    AVE = 1;
    STOCHASTIC = 2;
  }
  optional PoolMethod pool = 1 [default = MAX]; // The pooling method
  optional uint32 kernel_size = 2; // The kernel size
  optional uint32 stride = 3 [default = 1]; // The stride
  // The padding size -- currently implemented only for average pooling.
  optional uint32 pad = 4 [default = 0];
}

// Message that stores parameters used by PowerLayer
message PowerParameter {
  // PowerLayer computes outputs y = (shift + scale * x) ^ power.
  optional float power = 1 [default = 1.0];
  optional float scale = 2 [default = 1.0];
  optional float shift = 3 [default = 0.0];
}

// Message that stores parameters used by WindowDataLayer
message WindowDataParameter {
  // Specify the data source.
  optional string source = 1;
  // For data pre-processing, we can do simple scaling and subtracting the
  // data mean, if provided. Note that the mean subtraction is always carried
  // out before scaling.
  optional float scale = 2 [default = 1];
  optional string mean_file = 3;
  // Specify the batch size.
  optional uint32 batch_size = 4;
  // Specify if we would like to randomly crop an image.
  optional uint32 crop_size = 5 [default = 0];
  // Specify if we want to randomly mirror data.
  optional bool mirror = 6 [default = false];
  // Foreground (object) overlap threshold
  optional float fg_threshold = 7 [default = 0.5];
  // Background (non-object) overlap threshold
  optional float bg_threshold = 8 [default = 0.5];
  // Fraction of batch that should be foreground objects
  optional float fg_fraction = 9 [default = 0.25];
  // Amount of contextual padding to add around a window
  // (used only by the window_data_layer)
  optional uint32 context_pad = 10 [default = 0];
  // Mode for cropping out a detection window
  // warp: cropped window is warped to a fixed size and aspect ratio
  // square: the tightest square around the window is cropped
  optional string crop_mode = 11 [default = "warp"];
}

// DEPRECATED: V0LayerParameter is the old way of specifying layer parameters
// in Caffe. We keep this message type around for legacy support.
message V0LayerParameter {
  optional string name = 1; // the layer name
  optional string type = 2; // the string to specify the layer type

  // Parameters to specify layers with inner products.
  optional uint32 num_output = 3; // The number of outputs for the layer
  optional bool biasterm = 4 [default = true]; // whether to have bias terms
  optional FillerParameter weight_filler = 5; // The filler for the weight
  optional FillerParameter bias_filler = 6; // The filler for the bias

  optional uint32 pad = 7 [default = 0]; // The padding size
  optional uint32 kernelsize = 8; // The kernel size
  optional uint32 group = 9 [default = 1]; // The group size for group conv
  optional uint32 stride = 10 [default = 1]; // The stride
  enum PoolMethod {
    MAX = 0;
    AVE = 1;
    STOCHASTIC = 2;
  }
  optional PoolMethod pool = 11 [default = MAX]; // The pooling method
  optional float dropout_ratio = 12 [default = 0.5]; // dropout ratio

  optional uint32 local_size = 13 [default = 5]; // for local response norm
  optional float alpha = 14 [default = 1.]; // for local response norm
  optional float beta = 15 [default = 0.75]; // for local response norm

  // For data layers, specify the data source
  optional string source = 16;
  // For data pre-processing, we can do simple scaling and subtracting the
  // data mean, if provided. Note that the mean subtraction is always carried
  // out before scaling.
  optional float scale = 17 [default = 1];
  optional string meanfile = 18;
  // For data layers, specify the batch size.
  optional uint32 batchsize = 19;
  // For data layers, specify if we would like to randomly crop an image.
  optional uint32 cropsize = 20 [default = 0];
  // For data layers, specify if we want to randomly mirror data.
  optional bool mirror = 21 [default = false];

  // The blobs containing the numeric parameters of the layer
  repeated BlobProto blobs = 50;
  // The ratio that is multiplied on the global learning rate. If you want to
  // set the learning ratio for one blob, you need to set it for all blobs.
  repeated float blobs_lr = 51;
  // The weight decay that is multiplied on the global weight decay.
  repeated float weight_decay = 52;

  // The rand_skip variable is for the data layer to skip a few data points
  // to avoid all asynchronous sgd clients to start at the same point. The skip
  // point would be set as rand_skip * rand(0,1). Note that rand_skip should not
  // be larger than the number of keys in the leveldb.
  optional uint32 rand_skip = 53 [default = 0];

  // Fields related to detection (det_*)
  // foreground (object) overlap threshold
  optional float det_fg_threshold = 54 [default = 0.5];
  // background (non-object) overlap threshold
  optional float det_bg_threshold = 55 [default = 0.5];
  // Fraction of batch that should be foreground objects
  optional float det_fg_fraction = 56 [default = 0.25];

  // optional bool OBSOLETE_can_clobber = 57 [default = true];

  // Amount of contextual padding to add around a window
  // (used only by the window_data_layer)
  optional uint32 det_context_pad = 58 [default = 0];

  // Mode for cropping out a detection window
  // warp: cropped window is warped to a fixed size and aspect ratio
  // square: the tightest square around the window is cropped
  optional string det_crop_mode = 59 [default = "warp"];

  // For ReshapeLayer, one needs to specify the new dimensions.
  optional int32 new_num = 60 [default = 0];
  optional int32 new_channels = 61 [default = 0];
  optional int32 new_height = 62 [default = 0];
  optional int32 new_width = 63 [default = 0];

  // Whether or not ImageLayer should shuffle the list of files at every epoch.
  // It will also resize images if new_height or new_width are not zero.
  optional bool shuffle_images = 64 [default = false];

  // For ConcatLayer, one needs to specify the dimension for concatenation, and
  // the other dimensions must be the same for all the bottom blobs.
  // By default it will concatenate blobs along the channels dimension.
  optional uint32 concat_dim = 65 [default = 1];

  optional HDF5OutputParameter hdf5_output_param = 1001;
}