(Caffe,LeNet)初始化训练网络(三)

本文从CSDN上转移过来地址:
http://blog.csdn.net/mounty_fsc/article/details/51090306

1. Solver到Net

SGDSolver的构造函数中详见本系列博文(二),主要执行了其父类Solver的构造函数,接着执行Solver::Init()函数,在Init()中,有两个函数值得注意:InitTrainNet()InitTestNets()分别初始化训练网络和测试网络。

  • InitTrainNet
    • 首先,ReadNetParamsFromTextFileOrDie(param_.net(), &net_param)param_.net()(即examples/mnist/lenet_train_test.prototxt)中的信息读入net_param
    • 其次,net_.reset(new Net(net_param))重新构建网络,调用Net的构造方法。
    • 然后,在构造方法中执行Net::init(),开始正式创建网络。其主要代码如下:

template
void Net::Init(const NetParameter& in_param) {
...
for (int layer_id = 0; layer_id < param.layer_size(); ++layer_id) {

    // Setup layer.
    const LayerParameter& layer_param = param.layer(layer_id);
 
    layers_.push_back(LayerRegistry::CreateLayer(layer_param));

    // Figure out this layer's input and output
    for (int bottom_id = 0; bottom_id < layer_param.bottom_size();  ++bottom_id) {
      const int blob_id = AppendBottom(param, layer_id, bottom_id, &available_blobs, &blob_name_to_idx);
      // If a blob needs backward, this layer should provide it.
      need_backward |= blob_need_backward_[blob_id];
    }
    int num_top = layer_param.top_size();
    for (int top_id = 0; top_id < num_top; ++top_id) {
      AppendTop(param, layer_id, top_id, &available_blobs, &blob_name_to_idx);
    }
 ...

  layers_[layer_id]->SetUp(bottom_vecs_[layer_id], top_vecs_[layer_id]);
  ...
 }

for (int param_id = 0; param_id < num_param_blobs; ++param_id) {
  AppendParam(param, layer_id, param_id);
}

...
}
```

**说明:**

  1. Lenet5在caffe中共有9层,即`param.layer_size() == 5`,以上代码每一次for循环创建一个网络层
  2. 每层网络是通过`LayerRegistry::CreateLayer()`创建的,类似与Solver的创建(详见本系列博文(二))
  3. 14行`Net::AppendBottom()`,对于`layer_id`这层,从`Net::blob_`中取出blob放入该层对应的`bottom_vecs_[layer_id]`中
  4. 20行`Net::AppendTop()`,对于`layer_id`这层,创建`blob`(未包含数据)并放入`Net::blob_`中
  5. `Layer::SetUp()`
  6. `AppendParam`中把每层网络的训练参数与网络变量`learnable_params_`绑定,在lenet中,只有`conv1`,`conv2`,`ip1`,`ip2`四层有参数,每层分别有参数与偏置参数两项参数,因而`learnable_params_`的size为8.
  • InitTestNets
    该部分内容见本系列博文:(Caffe,Lenet5)初始化测试网络(四)

2 训练网络结构

Layer layer Type Bottom Blob Top Blob Top Blob Shape
1 minst Data data&&label 64 1 28 28 (50176) && 64 (64)
2 conv1 Convolution data conv1 64 20 24 24 (737280)
3 pool1 Pooling conv1 pool1 64 20 12 12 (184320)
4 conv2 Convolution pool1 conv2 64 50 8 8 (204800)
5 pool2 Pooling conv2 pool2 64 50 4 4 (51200)
6 ip1 InnerProduct pool2 ip1 64 500 (32000)
7 relu1 ReLU ip1 ip1(in-place) 64 500 (32000)
8 ip2 InnerProduct ip1 ip2 64 10 (640)
9 loss SoftmaxWithLoss ip2&&label loss (1)

注:Top Blob Shape格式为:BatchSize,ChannelSize,Height,Width(Total Count)

3 第一层:Data Layer

3.1 protobuff定义

训练网络的第一层protobuff定义为:

layer {
  name: "mnist"
  type: "Data"
  top: "data"
  top: "label"
  include {
    phase: TRAIN
  }
  transform_param {
    scale: 0.00390625
  }
  data_param {
    source: "examples/mnist/mnist_train_lmdb"
    batch_size: 64
    backend: LMDB
  }
}

3.2 函数LayerRegistry::CreateLayer

第1节中代码第一次通过调用LayerRegistry::CreateLayer()创建了DataLayer类,DataLayer类的继承关系如下图所示,详见[1]:

(Caffe,LeNet)初始化训练网络(三)_第1张图片

由继承图可知,调用DataLayer()的构造函数,依次执行的顺序为其基类构造函数:Layer()、BaseDataLayer()、InternalThread()(详见(Caffe)基本类InternalThread(三) )、BasePrefetchingDataLayer()、及DataLayer()

其中,值得注意的是DataLayer(),在调用基类构造函数BasePrefetchingDataLayer()之后,对 DataReader reader_ 进行赋值,在该DataLayer对象中维护了一个DataReader对象reader_,其作用是添加读取数据任务至,一个专门读取数据库(examples/mnist/mnist_train_lmdb)的线程(若还不存在该线程,则创建该线程),此处一共取出了4*64个样本至BlockingQueue DataReader::QueuePair::full_。详见(Caffe)基本类DataReader、QueuePair、Body(四)

template 
DataLayer::DataLayer(const LayerParameter& param)
  : BasePrefetchingDataLayer(param),
    reader_(param) {
}

3.3 函数Layer::SetUp

  • 此处按程序执行顺序值得关注的有:
    DataLayer::DataLayerSetUp中根据3.2DataReader中介绍的读取的数据中取出一个样本推测blob的形状

  • BasePrefetchingDataLayer::LayerSetUp如下代码prefetch_[i].data_.mutable_cpu_data()用到了涉及到gpu、cpu间复制数据的问题,见(Caffe)基本类Blob,Layer,Net(一)1.4SyncedMemory及引用[2]

     // Before starting the prefetch thread, we make cpu_data and gpu_data
     // calls so that the prefetch thread does not accidentally make simultaneous
     // cudaMalloc calls when the main thread is running. In some GPUs this
     // seems to cause failures if we do not so.
     for (int i = 0; i < PREFETCH_COUNT; ++i) {
       prefetch_[i].data_.mutable_cpu_data();
       if (this->output_labels_) {
         prefetch_[i].label_.mutable_cpu_data();
       }
     }
    
  • BasePrefetchingDataLayer类继承了InternalThread,BasePrefetchingDataLayer::LayerSetUp中通过调用StartInternalThread()开启了一个新线程,从而执行BasePrefetchingDataLayer::InternalThreadEntry

  • BasePrefetchingDataLayer::InternalThreadEntry关键代码如下,其中load_batch(batch)为,从2.2介绍的BlockingQueue DataReader::QueuePair::full_(包含从数据库读出的数据)中读取一个batch_size的数据到BlockingQueue*> BasePrefetchingDataLayer::prefetch_full_中。由于该线程在prefetch_free_为空时将挂起等待(PREFETCH_COUNT=3),prefetch_full_中用完的Batch将放回prefetch_free_中。该线程何时停止?

        while (!must_stop()) {
          Batch* batch = prefetch_free_.pop();
          load_batch(batch);
    #ifndef CPU_ONLY
          if (Caffe::mode() == Caffe::GPU) {
            batch->data_.data().get()->async_gpu_push(stream);
            CUDA_CHECK(cudaStreamSynchronize(stream));
          }
    #endif
          prefetch_full_.push(batch);
        }
    

关于线程的总结

  1. 此外一共涉及到两个线程,分别为都是继承了InnerThreadBasePrefetchingDataLayer(DataLayer)类和DataReader中的Body
  2. Body为面向数据库的线程,不断从某个数据库中读出数据,存放至缓存为队列DataReader::QueuePair::BlockingQueue,一般保存4*64个单位数据,单位为Datum
  3. BasePrefetchingDataLayer为面向网络的线程,从Body的缓存中不断读取数据。BasePrefetchingDataLayer的缓存为队列BlockingQueue,一般存放3个单位的数据,单位为Batch
static const int PREFETCH_COUNT = 3;
Batch prefetch_[PREFETCH_COUNT];
BlockingQueue*> prefetch_free_;
BlockingQueue*> prefetch_full_;

template 
BasePrefetchingDataLayer::BasePrefetchingDataLayer(
    const LayerParameter& param)
    : BaseDataLayer(param),
      prefetch_free_(), prefetch_full_() {
  for (int i = 0; i < PREFETCH_COUNT; ++i) {
    prefetch_free_.push(&prefetch_[i]);
  }
}
  • prefetch_full_prefetch_free_中的元素由prefetch_提供

4 第二层:Convolution Layer

4.1 protobuff定义

layer {
  name: "conv1"
  type: "Convolution"
  bottom: "data"
  top: "conv1"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  convolution_param {
    num_output: 20
    kernel_size: 5
    stride: 1
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
    }
  }
}

4.2 函数LayerRegistry::CreateLayer

(Caffe,LeNet)初始化训练网络(三)_第2张图片
这里写图片描述

说明:

  1. 不像DataLayer 直接执行的是构造函数,此时执行的是GetConvolutuionLayer(),然后调用ConvolutionLayer(),原因如下:

REGISTER_LAYER_CREATOR(Convolution, GetConvolutionLayer);

4.3 Layer::SetUp

Layer::SetUp中,调用了ConvolutionLayer的基类BaseConvolutionLayerLayerSetUp及Reshape函数,该类的主要成员变量如下:

/**
 * @brief Abstract base class that factors out the BLAS code common to
 *        ConvolutionLayer and DeconvolutionLayer.
 */
template 
class BaseConvolutionLayer : public Layer {
 public:
  explicit BaseConvolutionLayer(const LayerParameter& param)
      : Layer(param) {}
  virtual void LayerSetUp(const vector*>& bottom,
      const vector*>& top);
  virtual void Reshape(const vector*>& bottom,
      const vector*>& top);

 ...
  /// @brief The spatial dimensions of a filter kernel.
  Blob kernel_shape_;
  /// @brief The spatial dimensions of the stride.
  Blob stride_;
  /// @brief The spatial dimensions of the padding.
  Blob pad_;
  /// @brief The spatial dimensions of the dilation.
  Blob dilation_;
  /// @brief The spatial dimensions of the convolution input.
  Blob conv_input_shape_;
  /// @brief The spatial dimensions of the col_buffer.
  vector col_buffer_shape_;
  /// @brief The spatial dimensions of the output.
  vector output_shape_;
  const vector* bottom_shape_;
...
};

说明:

  1. LayerSetUp函数中,主要是初始化了kernel_shape_、stride_、pad_、dilation_以及初始化网络参数,并存放与Layer::blobs_中。
  2. Reshape函数中,conv_input_shape_、bottom_shape_等

5 第三层:Pooling Layer

5.1 protobuff定义

layer {
  name: "pool1"
  type: "Pooling"
  bottom: "conv1"
  top: "pool1"
  pooling_param {
    pool: MAX
    kernel_size: 2
    stride: 2
  }
}

5.2 Layer::SetUp

通过调用虚函数LayerSetUpReshape对以下成员变量进行初始化

/**
 * @brief Pools the input image by taking the max, average, etc. within regions.
 *
 * TODO(dox): thorough documentation for Forward, Backward, and proto params.
 */
template 
class PoolingLayer : public Layer {
 ....
  int kernel_h_, kernel_w_;
  int stride_h_, stride_w_;
  int pad_h_, pad_w_;
  int channels_;
  int height_, width_;
  int pooled_height_, pooled_width_;
  bool global_pooling_;
  Blob rand_idx_;
  Blob max_idx_;
};

6 第四层、第五层

基本同第二层、第三层

7 第六层:InnerProduct Layer

7.1 protobuff定义

layer {
  name: "ip1"
  type: "InnerProduct"
  bottom: "pool2"
  top: "ip1"
  param {
    lr_mult: 1
  }
  param {
    lr_mult: 2
  }
  inner_product_param {
    num_output: 500
    weight_filler {
      type: "xavier"
    }
    bias_filler {
      type: "constant"
    }
  }
}

7.2 Layer::SetUp

/**
 * @brief Also known as a "fully-connected" layer, computes an inner product
 *        with a set of learned weights, and (optionally) adds biases.
 *
 * TODO(dox): thorough documentation for Forward, Backward, and proto params.
 */
template 
class InnerProductLayer : public Layer {
 ...
  int M_;
  int K_;
  int N_;
  bool bias_term_;
  Blob bias_multiplier_;
};

说明:

  1. N_为输出大小,即等于protobuff中定义的num_output
  2. K_为输入大小,对于该层Bottom Blob形状为(N, C, H, W),N为batch_size,K_=CHW(Caffe)基本类Blob,Layer,Net(一),M_=N。其中只有C、H、W跟内积相关

8 第七层:ReLU Layer

8.1 protobuff定义

layer {
  name: "relu1"
  type: "ReLU"
  bottom: "ip1"
  top: "ip1"
}

8.2 说明

ReLULayer主要是用来做计算的,其继承关系如下,详细参加[4]、[5]

(Caffe,LeNet)初始化训练网络(三)_第3张图片

9 第八层:InnerProduct Layer

参见第7节

10 第九层:SoftmaxWithLoss Layer

10.1 protobuff定义

layer {
  name: "loss"
  type: "SoftmaxWithLoss"
  bottom: "ip2"
  bottom: "label"
  top: "loss"
}

10.2 LayerRegistry::CreateLayer

(Caffe,LeNet)初始化训练网络(三)_第4张图片
这里写图片描述

10.3 Layer::SetUp

值得注意的是:

  1. SoftmaxWithLossLayer包含类SoftmaxLayer的实例
    shared_ptr > softmax_layer_

  2. softmax_layer_LayerSetUp中赋值。

  3. 此函数内调用Layer::SetLossWeights初始化了该层的Top Blob(loss)

  4. 两个类间的关系如下图:

    (Caffe,LeNet)初始化训练网络(三)_第5张图片
    这里写图片描述
  5. 成员变量prob_作为Softmaxlayer的top blob

  6. bottom blob[0]作为softmaxlayer的bottom blob

  7. 所以经过softmaxlayer计算之后,得出64*10(每个样本的每个类别上的概率)存放在prob_中

11 剩余的工作

至此,训练网络基本创建完毕,接下来剩下的工作主要有:

  1. 反向检查一次网络,看哪些blobs会对loss产生影响,在LeNet5中,前面的9层均有影响
  2. 初始化权值共享

[1].http://caffe.berkeleyvision.org/doxygen/classcaffe_1_1BasePrefetchingDataLayer.html
[2].http://caffe.berkeleyvision.org/tutorial/net_layer_blob.html Implementation Details
[3].http://caffe.berkeleyvision.org/doxygen/classcaffe_1_1ConvolutionLayer.html
[4].http://caffe.berkeleyvision.org/doxygen/classcaffe_1_1ReLULayer.html
[5].http://caffe.berkeleyvision.org/tutorial/layers.html ReLU / Rectified-Linear and Leaky-ReLU

你可能感兴趣的:((Caffe,LeNet)初始化训练网络(三))