caffe源碼解讀(13)-blob.hpp

　　caffe使用稱爲blob的四維數組用於存儲和交換數據。Blob提供了統一的存儲器接口，持有一批圖像或其他數據、權值、權值更新值。Blob在內存中表示四維數組，維度從低到高爲[width_,height_,channels_,num_]，其中，width_和height_表示圖像的寬和高，channes_表示顏色通道RGB，num_表示第幾幀，用於存儲數據或權值(data)和權值增量(diff)，在進行網絡計算時，每層的輸入和輸出都需要通過Blob對象緩衝。Blob是caffe的基本存儲單元，下面對其進行簡單介紹，主要參考趙永科所著的《深度學習.21天實戰caffe》中對該Blob的詳細介紹。
　　數據結構描述
　　打開src/caffe/proto/caffe.proto，首先映入眼簾的便是與Blob相關的描述，可見該數據結構的重要性，是其他大部分數據結構的依賴項。

//該結構描述了Blob的形狀信息
message BlobShape {
  repeated int64 dim = 1 [packed = true];
  //只包括若干int64類型值，分別表示Blob每個維度大小，
  //packed表示這些值在內存中緊密排列，沒有空洞
}
//該結構描述了Blob在磁盤中序列化後的狀態
message BlobProto {
　//可選，包括一個BlobShape對象
  optional BlobShape shape = 7;
  //包括若干浮點元素，存儲數據或權值
  //元素數目由shape或(num,channel,height,width)確定
  //這些元素在內存中緊密排列
  repeated float data = 5 [packed = true];
  repeated float diff = 6 [packed = true];
  repeated double double_data = 8 [packed = true];//與data並列，只是類型爲double
  repeated double double_diff = 9 [packed = true];//與diff並列，只是類型爲double

  // 4D dimensions -- deprecated.  Use "shape" instead.
  //以下爲可選的維度信息，新版本caffe推薦使用shape，而不再用後面的值
  optional int32 num = 1 [default = 0];
  optional int32 channels = 2 [default = 0];
  optional int32 height = 3 [default = 0];
  optional int32 width = 4 [default = 0];
}

Blob是怎樣練成的
Blob是一個模板類，聲明在include/caffe/blob.hpp中，封裝了SynceMemory。

#ifndef CAFFE_BLOB_HPP_
#define CAFFE_BLOB_HPP_
#include <algorithm>
#include <string>
#include <vector>
#include "caffe/common.hpp"
//由protoc生成的頭文件
#include "caffe/proto/caffe.pb.h"
//CPU/GPU共享內存類，用於數據同步
#include "caffe/syncedmem.hpp"

//Blob最大維數目
const int kMaxBlobAxes = 32;
namespace caffe {
template <typename Dtype>
class Blob {//類聲明
 public:
 //默認構造函數
  Blob()
       : data_(), diff_(), count_(0), capacity_(0) {}

  explicit Blob(const int num, const int channels, const int height,
  //顯式構造函數，避免隱式數據類型轉換
      const int width);
  explicit Blob(const vector<int>& shape);

  void Reshape(const int num, const int channels, const int height,
      const int width);
  //變形函數，根據輸入參數重新設置當前blob形狀，必要時重新分配內存
  void Reshape(const vector<int>& shape);
  void Reshape(const BlobShape& shape);
  void ReshapeLike(const Blob& other);
  //得到Blob形狀字符串，用於打印log,見caffe運行log，
  //類似於"Top shape: 100 1 28 28 (78400)"
  inline string shape_string() const {
    ostringstream stream;
    for (int i = 0; i < shape_.size(); ++i) {
      stream << shape_[i] << " ";
    }
    stream << "(" << count_ << ")";
    return stream.str();
  }
  //返回Blob形狀
  inline const vector<int>& shape() const { return shape_; }
  //返回某一維度的尺寸
  inline int shape(int index) const {
    return shape_[CanonicalAxisIndex(index)];
  }
  //返回維度數目
  inline int num_axes() const { return shape_.size(); }
  //返回Blob中元素總數
  inline int count() const { return count_; }
　//返回Blob中某幾維子集的元素總數
  inline int count(int start_axis, int end_axis) const {
    CHECK_LE(start_axis, end_axis);//保證start_axis<=end_axis
    CHECK_GE(start_axis, 0);//保證start_axis>= 0
    CHECK_GE(end_axis, 0);//保證end_axis>= 0
    CHECK_LE(start_axis, num_axes());//保證start_axis<=總的維度數目
    CHECK_LE(end_axis, num_axes());//保證end_axis<=總的維度數目
    int count = 1;
    for (int i = start_axis; i < end_axis; ++i) {
      count *= shape(i);
    }
    return count;
  }
  //計算從某一維度開始的元素總數
  inline int count(int start_axis) const {
    return count(start_axis, num_axes());
  }
//轉換座標軸索引[-N,N]爲普通索引[0,N]
  inline int CanonicalAxisIndex(int axis_index) const {
    CHECK_GE(axis_index, -num_axes())//保證axis_index>=-num_axes()
        << "axis " << axis_index << " out of range for " << num_axes()
        << "-D Blob with shape " << shape_string();
    CHECK_LT(axis_index, num_axes())//保證axis_index<num_axes()
        << "axis " << axis_index << " out of range for " << num_axes()
        << "-D Blob with shape " << shape_string();
    if (axis_index < 0) {
      return axis_index + num_axes();
    }
    return axis_index;
  }
  //獲取形狀某一維的尺寸
  inline int num() const { return LegacyShape(0); }
  inline int channels() const { return LegacyShape(1); }
  inline int height() const { return LegacyShape(2); 
  inline int width() const { return LegacyShape(3); }
  inline int LegacyShape(int index) const {
    CHECK_LE(num_axes(), 4)
        << "Cannot use legacy accessors on Blobs with > 4 axes.";
    CHECK_LT(index, 4);
    CHECK_GE(index, -4);
    if (index >= num_axes() || index < -num_axes()) {
      return 1;
    }
    return shape(index);
  }
//下面這幾個函數都是計算偏移量的
  inline int offset(const int n, const int c = 0, const int h = 0,
      const int w = 0) const {
    CHECK_GE(n, 0);
    CHECK_LE(n, num());
    CHECK_GE(channels(), 0);
    CHECK_LE(c, channels());
    CHECK_GE(height(), 0);
    CHECK_LE(h, height());
    CHECK_GE(width(), 0);
    CHECK_LE(w, width());
    return ((n * channels() + c) * height() + h) * width() + w;
  }

  inline int offset(const vector<int>& indices) const {
    CHECK_LE(indices.size(), num_axes());
    int offset = 0;
    for (int i = 0; i < num_axes(); ++i) {
      offset *= shape(i);
      if (indices.size() > i) {
        CHECK_GE(indices[i], 0);
        CHECK_LT(indices[i], shape(i));
        offset += indices[i];
      }
    }
    return offset;
  }
  //按值拷貝Blob到當前Blob
  void CopyFrom(const Blob<Dtype>& source, bool copy_diff = false,
      bool reshape = false);

  inline Dtype data_at(const int n, const int c, const int h,
      const int w) const {
    return cpu_data()[offset(n, c, h, w)];
  }

  inline Dtype diff_at(const int n, const int c, const int h,
      const int w) const {
    return cpu_diff()[offset(n, c, h, w)];
  }

  inline Dtype data_at(const vector<int>& index) const {
    return cpu_data()[offset(index)];
  }

  inline Dtype diff_at(const vector<int>& index) const {
    return cpu_diff()[offset(index)];
  }

  inline const shared_ptr<SyncedMemory>& data() const {
    CHECK(data_);
    return data_;
  }

  inline const shared_ptr<SyncedMemory>& diff() const {
    CHECK(diff_);
    return diff_;
  }
  //只讀訪問cpu data
  const Dtype* cpu_data() const;
  //設置cpu data
  void set_cpu_data(Dtype* data);
  //只讀訪問gpu data
  const int* gpu_shape() const;
  const Dtype* gpu_data() const;
  //設置gpu data
  void set_gpu_data(Dtype* data);
  //只讀訪問cpu diff
  const Dtype* cpu_diff() const;
  //只讀訪問gpu diff
  const Dtype* gpu_diff() const;
  //讀寫訪問cpu data、gpu data、cpu diff、gpu diff
  Dtype* mutable_cpu_data();
  Dtype* mutable_gpu_data();
  Dtype* mutable_cpu_diff();
  Dtype* mutable_gpu_diff();
  void Update();//Blob更新運算，簡單理解爲data與diff的merge過程
  //反序列化函數，從BlobProto中回覆一個Blob對象
  void FromProto(const BlobProto& proto, bool reshape = true);
  //序列化函數，將內存中的Blob對象保存到BlobProto中
  void ToProto(BlobProto* proto, bool write_diff = false) const;
　//計算data的L1範數（qiuhe                    ）
  Dtype asum_data() const;
  //計算diff的L1範數
  Dtype asum_diff() const;
 //計算data的L2範數
  Dtype sumsq_data() const;
　//計算diff的L2範數
  Dtype sumsq_diff() const;

  void scale_data(Dtype scale_factor);
  void scale_diff(Dtype scale_factor);
  void ShareData(const Blob& other);
  void ShareDiff(const Blob& other);
  bool ShapeEquals(const BlobProto& other);

 protected:
  shared_ptr<SyncedMemory> data_;//存放指向data的指針
  shared_ptr<SyncedMemory> diff_;//存放指向diff的指針
  shared_ptr<SyncedMemory> shape_data_;
  vector<int> shape_;//形狀信息
  int count_;//存放有效元素數目信息
  int capacity_;//存放Blob容器的容量信息
  DISABLE_COPY_AND_ASSIGN(Blob);
};  // class Blob

}  // namespace caffe

#endif  // CAFFE_BLOB_HPP_