caffe-----使用C++ 提取网络中间层特征数据

最近实验，想要在c++下知道网络中间某一层的特征数据情况，查找了相关资料，记录一下。

其实在caffe框架里面是包含这种操作的，可以模仿tools/extract_features.cpp中的操作来得到网络中间的特征数据。

首先看下extract_features.cpp是如何写的。

template<typename Dtype>
int feature_extraction_pipeline(int argc, char** argv) {
  ::google::InitGoogleLogging(argv[0]);
  const int num_required_args = 7;
  if (argc < num_required_args) {
    LOG(ERROR)<<
    "This program takes in a trained network and an input data layer, and then"
    " extract features of the input data produced by the net.\n"
    "Usage: extract_features  pretrained_net_param"
    "  feature_extraction_proto_file  extract_feature_blob_name1[,name2,...]"
    "  save_feature_dataset_name1[,name2,...]  num_mini_batches  db_type"
    "  [CPU/GPU] [DEVICE_ID=0]\n"
    "Note: you can extract multiple features in one pass by specifying"
    " multiple feature blob names and dataset names separated by ','."
    " The names cannot contain white space characters and the number of blobs"
    " and datasets must be equal.";
    return 1;
  }
  int arg_pos = num_required_args;

  arg_pos = num_required_args;
  if (argc > arg_pos && strcmp(argv[arg_pos], "GPU") == 0) {
    LOG(ERROR)<< "Using GPU";
    int device_id = 0;
    if (argc > arg_pos + 1) {
      device_id = atoi(argv[arg_pos + 1]);
      CHECK_GE(device_id, 0);
    }
    LOG(ERROR) << "Using Device_id=" << device_id;
    Caffe::SetDevice(device_id);
    Caffe::set_mode(Caffe::GPU);
  } else {
    LOG(ERROR) << "Using CPU";
    Caffe::set_mode(Caffe::CPU);
  }

  arg_pos = 0;  // the name of the executable
  std::string pretrained_binary_proto(argv[++arg_pos]);

  // Expected prototxt contains at least one data layer such as
  //  the layer data_layer_name and one feature blob such as the
  //  fc7 top blob to extract features.
  /*
   layers {
     name: "data_layer_name"
     type: DATA
     data_param {
       source: "/path/to/your/images/to/extract/feature/images_leveldb"
       mean_file: "/path/to/your/image_mean.binaryproto"
       batch_size: 128
       crop_size: 227
       mirror: false
     }
     top: "data_blob_name"
     top: "label_blob_name"
   }
   layers {
     name: "drop7"
     type: DROPOUT
     dropout_param {
       dropout_ratio: 0.5
     }
     bottom: "fc7"
     top: "fc7"
   }
   */
  std::string feature_extraction_proto(argv[++arg_pos]);
  boost::shared_ptr<Net<Dtype> > feature_extraction_net(
      new Net<Dtype>(feature_extraction_proto, caffe::TEST));
  feature_extraction_net->CopyTrainedLayersFrom(pretrained_binary_proto);//初始化网络

  std::string extract_feature_blob_names(argv[++arg_pos]);
  std::vector<std::string> blob_names;
  boost::split(blob_names, extract_feature_blob_names, boost::is_any_of(","));

  std::string save_feature_dataset_names(argv[++arg_pos]);
  std::vector<std::string> dataset_names;
  boost::split(dataset_names, save_feature_dataset_names,
               boost::is_any_of(","));
  CHECK_EQ(blob_names.size(), dataset_names.size()) <<
      " the number of blob names and dataset names must be equal";
  size_t num_features = blob_names.size();

  for (size_t i = 0; i < num_features; i++) {
    CHECK(feature_extraction_net->has_blob(blob_names[i]))
        << "Unknown feature blob name " << blob_names[i]
        << " in the network " << feature_extraction_proto;
  }

  int num_mini_batches = atoi(argv[++arg_pos]);

  std::vector<boost::shared_ptr<db::DB> > feature_dbs;
  std::vector<boost::shared_ptr<db::Transaction> > txns;
  const char* db_type = argv[++arg_pos];
  for (size_t i = 0; i < num_features; ++i) {
    LOG(INFO)<< "Opening dataset " << dataset_names[i];
    boost::shared_ptr<db::DB> db(db::GetDB(db_type));
    db->Open(dataset_names.at(i), db::NEW);
    feature_dbs.push_back(db);
    boost::shared_ptr<db::Transaction> txn(db->NewTransaction());
    txns.push_back(txn);
  }

  LOG(ERROR)<< "Extracting Features";

  Datum datum;
  std::vector<int> image_indices(num_features, 0);
  for (int batch_index = 0; batch_index < num_mini_batches; ++batch_index) {
    feature_extraction_net->Forward();//首先进行前传 这样才能有中间数据
    for (int i = 0; i < num_features; ++i) {
      const boost::shared_ptr<Blob<Dtype> > feature_blob =
        feature_extraction_net->blob_by_name(blob_names[i]);//通过名字查找blob
      int batch_size = feature_blob->num();
      int dim_features = feature_blob->count() / batch_size;
      const Dtype* feature_blob_data;
      for (int n = 0; n < batch_size; ++n) {
        datum.set_height(feature_blob->height());
        datum.set_width(feature_blob->width());
        datum.set_channels(feature_blob->channels());
        datum.clear_data();
        datum.clear_float_data();
        feature_blob_data = feature_blob->cpu_data() +
            feature_blob->offset(n);
        for (int d = 0; d < dim_features; ++d) {
          datum.add_float_data(feature_blob_data[d]);//将feature_blob的数据都保存到datum里
        }
        string key_str = caffe::format_int(image_indices[i], 10);

        string out;
        CHECK(datum.SerializeToString(&out));//将datum保存到本地
        txns.at(i)->Put(key_str, out);
        ++image_indices[i];
        if (image_indices[i] % 1000 == 0) {
          txns.at(i)->Commit();
          txns.at(i).reset(feature_dbs.at(i)->NewTransaction());
          LOG(ERROR)<< "Extracted features of " << image_indices[i] <<
              " query images for feature blob " << blob_names[i];
        }
      }  // for (int n = 0; n < batch_size; ++n)
    }  // for (int i = 0; i < num_features; ++i)
  }  // for (int batch_index = 0; batch_index < num_mini_batches; ++batch_index)
  // write the last batch
  for (int i = 0; i < num_features; ++i) {
    if (image_indices[i] % 1000 != 0) {
      txns.at(i)->Commit();
    }
    LOG(ERROR)<< "Extracted features of " << image_indices[i] <<
        " query images for feature blob " << blob_names[i];
    feature_dbs.at(i)->Close();
  }

  LOG(ERROR)<< "Successfully extracted the features!";
  return 0;
}

 

主要三个核心步骤：

1.初始化网络，并前传，

net->Forward()

2.通过blob的名字（prototxt中的name）来得到blob数据，

const boost::shared_ptr<Blob<Dtype> > feature_blob = net->blob_by_name(blob_names[i])

3.blob里面已经保存了所有的特征数据，按照需求取出来就好了。

count = feature_blob->channels() * feature_blob->height() *  feature_blob->width();
float* feature_array = new float[count]; 
const float* feature_blob_data = feature_blob->cpu_data() + feature_blob->offset(n); // feature data generated from 
                             // the nth input image within a batch 
memcpy(feature_array, feature_blob_data, count * sizeof(float)); 
...// other operations
delete [] feature_array;  

 

 

如下是做实验时候的一个例子，提取出了blstm_input中的数据，并保存到了txt里。

 

Blob<float>* input_layer = m_net->input_blobs()[0];
  input_layer->Reshape(1, m_channelNum, m_inputGeometry.height, m_inputGeometry.width);
  m_net->Reshape();
  std::vector<cv::Mat> input_channels;
  wrapInputLayer(&input_channels);
  preprocess(img, &input_channels);
  m_net->Forward();
  Blob<float>* output_layer = m_net->output_blobs()[0];  
  int alphabet_size=output_layer->shape(2);
  int time_step=output_layer->shape(0);

 vector<int> shape;

const boost::shared_ptr<Blob<float> > blstm_input = m_net->blob_by_name("blstm_input");
    shape = blstm_input->shape();
    for(int i = 0; i < shape.size(); i++)
    {
        cout<<" blstm_input shape:"<<i<<"    :"<<shape[i]<<endl;
    }
    
    const boost::shared_ptr<Blob<float> > lstm1 = m_net->blob_by_name("lstm1");
    shape = lstm1->shape();
    for(int i = 0; i < shape.size(); i++)
    {
        cout<<" lstm1 shape:"<<i<<"    :"<<shape[i]<<endl;
    }

    cout<<"==============blob info======="<<endl;
    ofstream of("blstm.txt");
    for(int h = 0; h < 192; h++)
    {
        int count = blstm_input->channels() * blstm_input->height() * blstm_input->width();
        // cout<<"blstm_input->channels():"<<blstm_input->channels()<<"    blstm_input->height():"<<blstm_input->height()
        //     <<"    blstm_input->width():"<<blstm_input->width()<<endl;
        float* feature_array = new float[count]; 
        const float* feature_blob_data = blstm_input->cpu_data() +
            blstm_input->offset(h); // feature data generated from the nth input image within a batch 
        memcpy(feature_array, feature_blob_data, count * sizeof(float)); 
        
        
        for(int i = 0; i < count; i++ )
        {
            if(i && i % 512 == 0)
            {
                of<<endl;
            }
            of<<"    ["<< h<< ","<<i % 512<< "]:"<<feature_blob_data[i];
        }
        of<<endl;
        delete [] feature_array;
    }
    of.close();

 

 

 

参考：

https://stackoverflow.com/questions/40938372/how-to-get-features-from-several-layers-using-c-in-caffe

转载于:https://www.cnblogs.com/hellowooorld/p/11348440.html