1. Porting Caffe onto ARM Compute Library.

2. The release version is 0.2.0

2 files changed, 1092 insertions, 0 deletions

diff --git a/examples/cpp_classification/classification_profiling.cpp b/examples/cpp_classification/classification_profiling.cpp
new file mode 100644
index 00000000..f5d5eaed
--- /dev/null
+++ b/examples/cpp_classification/classification_profiling.cpp
@@ -0,0 +1,546 @@
+#include <caffe/caffe.hpp>
+#ifdef USE_OPENCV
+#include <opencv2/core/core.hpp>
+#include <opencv2/highgui/highgui.hpp>
+#include <opencv2/imgproc/imgproc.hpp>
+#endif  // USE_OPENCV
+#include <algorithm>
+#include <iosfwd>
+#include <memory>
+#include <string>
+#include <utility>
+#include <vector>
+
+#ifdef USE_PROFILING
+
+#include <iostream>
+
+#include <time.h>
+
+#define REPEAT_TEST
+
+unsigned long get_cur_time(void)
+{
+   struct timespec tm;
+
+   clock_gettime(CLOCK_MONOTONIC_COARSE, &tm);
+
+   return (tm.tv_sec*1000+tm.tv_nsec/1000000);
+}
+
+#endif //USE_PROFILING
+
+#ifdef USE_OPENCV
+using namespace caffe;  // NOLINT(build/namespaces)
+using std::string;
+
+/* Pair (label, confidence) representing a prediction. */
+typedef std::pair<string, float> Prediction;
+
+class Classifier {
+ public:
+  Classifier(const string& model_file,
+             const string& trained_file,
+             const string& mean_file,
+             const string& label_file);
+
+  std::vector<Prediction> Classify(const cv::Mat& img, int N = 5);
+
+#ifdef USE_PROFILING
+
+#ifdef LAYER_PERF_STAT
+  void  dump_perf_stat(void);
+  void  dump_single_layer_io(int idx, Layer<float> * p_layer);
+  void  dump_single_layer_perf(int idx, Layer<float> * p_layer,uint64_t total_net_time);
+#ifdef REPEAT_TEST
+  void collect_layer_stat(vector<vector<perf_stat> * > & all_stat);
+  void dump_all_stat(vector <vector<perf_stat>*>& all_stat);
+  void reset_layer_stat();
+#endif
+#endif
+
+#endif //USE_PROFILING
+
+ private:
+  void SetMean(const string& mean_file);
+
+  std::vector<float> Predict(const cv::Mat& img);
+
+  void WrapInputLayer(std::vector<cv::Mat>* input_channels);
+
+  void Preprocess(const cv::Mat& img,
+                  std::vector<cv::Mat>* input_channels);
+
+ private:
+  shared_ptr<Net<float> > net_;
+  cv::Size input_geometry_;
+  int num_channels_;
+  cv::Mat mean_;
+  std::vector<string> labels_;
+};
+
+Classifier::Classifier(const string& model_file,
+                       const string& trained_file,
+                       const string& mean_file,
+                       const string& label_file) {
+#ifdef CPU_ONLY
+  Caffe::set_mode(Caffe::CPU);
+#else
+  Caffe::set_mode(Caffe::GPU);
+#endif
+
+  /* Load the network. */
+  net_.reset(new Net<float>(model_file, TEST));
+  net_->CopyTrainedLayersFrom(trained_file);
+
+  CHECK_EQ(net_->num_inputs(), 1) << "Network should have exactly one input.";
+  CHECK_EQ(net_->num_outputs(), 1) << "Network should have exactly one output.";
+
+  Blob<float>* input_layer = net_->input_blobs()[0];
+  num_channels_ = input_layer->channels();
+  CHECK(num_channels_ == 3 || num_channels_ == 1)
+    << "Input layer should have 1 or 3 channels.";
+  input_geometry_ = cv::Size(input_layer->width(), input_layer->height());
+
+  /* Load the binaryproto mean file. */
+  SetMean(mean_file);
+
+  /* Load labels. */
+  std::ifstream labels(label_file.c_str());
+  CHECK(labels) << "Unable to open labels file " << label_file;
+  string line;
+  while (std::getline(labels, line))
+    labels_.push_back(string(line));
+
+  Blob<float>* output_layer = net_->output_blobs()[0];
+  CHECK_EQ(labels_.size(), output_layer->channels())
+    << "Number of labels is different from the output layer dimension.";
+}
+
+static bool PairCompare(const std::pair<float, int>& lhs,
+                        const std::pair<float, int>& rhs) {
+  return lhs.first > rhs.first;
+}
+
+/* Return the indices of the top N values of vector v. */
+static std::vector<int> Argmax(const std::vector<float>& v, int N) {
+  std::vector<std::pair<float, int> > pairs;
+  for (size_t i = 0; i < v.size(); ++i)
+    pairs.push_back(std::make_pair(v[i], i));
+  std::partial_sort(pairs.begin(), pairs.begin() + N, pairs.end(), PairCompare);
+
+  std::vector<int> result;
+  for (int i = 0; i < N; ++i)
+    result.push_back(pairs[i].second);
+  return result;
+}
+
+/* Return the top N predictions. */
+std::vector<Prediction> Classifier::Classify(const cv::Mat& img, int N) {
+  std::vector<float> output = Predict(img);
+
+  N = std::min<int>(labels_.size(), N);
+  std::vector<int> maxN = Argmax(output, N);
+  std::vector<Prediction> predictions;
+  for (int i = 0; i < N; ++i) {
+    int idx = maxN[i];
+    predictions.push_back(std::make_pair(labels_[idx], output[idx]));
+  }
+
+  return predictions;
+}
+
+/* Load the mean file in binaryproto format. */
+void Classifier::SetMean(const string& mean_file) {
+  BlobProto blob_proto;
+  ReadProtoFromBinaryFileOrDie(mean_file.c_str(), &blob_proto);
+
+  /* Convert from BlobProto to Blob<float> */
+  Blob<float> mean_blob;
+  mean_blob.FromProto(blob_proto);
+  CHECK_EQ(mean_blob.channels(), num_channels_)
+    << "Number of channels of mean file doesn't match input layer.";
+
+  /* The format of the mean file is planar 32-bit float BGR or grayscale. */
+  std::vector<cv::Mat> channels;
+  float* data = mean_blob.mutable_cpu_data();
+  for (int i = 0; i < num_channels_; ++i) {
+    /* Extract an individual channel. */
+    cv::Mat channel(mean_blob.height(), mean_blob.width(), CV_32FC1, data);
+    channels.push_back(channel);
+    data += mean_blob.height() * mean_blob.width();
+  }
+
+  /* Merge the separate channels into a single image. */
+  cv::Mat mean;
+  cv::merge(channels, mean);
+
+  /* Compute the global mean pixel value and create a mean image
+   * filled with this value. */
+  cv::Scalar channel_mean = cv::mean(mean);
+  mean_ = cv::Mat(input_geometry_, mean.type(), channel_mean);
+}
+
+std::vector<float> Classifier::Predict(const cv::Mat& img) {
+  Blob<float>* input_layer = net_->input_blobs()[0];
+  input_layer->Reshape(1, num_channels_,
+                       input_geometry_.height, input_geometry_.width);
+  /* Forward dimension change to all layers. */
+  net_->Reshape();
+
+  std::vector<cv::Mat> input_channels;
+  WrapInputLayer(&input_channels);
+
+  Preprocess(img, &input_channels);
+
+#ifdef USE_PROFILING
+  unsigned long tstart=get_cur_time();
+#endif //USE_PROFILING
+
+  net_->Forward();
+
+#ifdef USE_PROFILING
+
+  unsigned long tend=get_cur_time();
+
+  std::cout<<"used time: "<<tend-tstart<<std::endl;
+
+#ifdef LAYER_PERF_STAT
+  dump_perf_stat(); 
+#ifdef REPEAT_TEST
+
+   reset_layer_stat();
+
+   vector<vector<perf_stat>* >  all_stat;
+   int rep_number=10;
+
+   for(int i=0;i<rep_number;i++)
+   {
+      net_->Forward();
+      collect_layer_stat(all_stat);
+      reset_layer_stat();
+   }
+
+   //dump stats
+   dump_all_stat(all_stat);
+
+   for(int i=0;i<all_stat.size();i++)
+         delete all_stat[i];
+   
+#endif //REPEAT_TEST
+#endif //LAYER_PERF_STAT
+#endif //USE_PROFILING
+
+  /* Copy the output layer to a std::vector */
+  Blob<float>* output_layer = net_->output_blobs()[0];
+  const float* begin = output_layer->cpu_data();
+  const float* end = begin + output_layer->channels();
+  return std::vector<float>(begin, end);
+}
+
+#ifdef USE_PROFILING
+
+#ifdef LAYER_PERF_STAT
+
+#ifdef REPEAT_TEST
+void Classifier::collect_layer_stat(vector<vector<perf_stat>*>& all_stat)
+{
+   vector<perf_stat > * p_stat;
+   perf_stat * p_time_stat;
+   const vector<shared_ptr<Layer<float> > >& layers=net_->layers();
+
+   
+   p_stat=new vector<perf_stat>;
+
+   for (int i =0;i< layers.size(); i++) {
+        p_time_stat=layers[i]->get_time_stat();
+        p_stat->push_back(*p_time_stat);
+
+   }
+
+   all_stat.push_back(p_stat);
+}
+
+void Classifier::reset_layer_stat(void)
+{
+   const vector<shared_ptr<Layer<float> > >& layers=net_->layers();
+   perf_stat * p_time_stat;
+
+   for (int i =0;i< layers.size(); i++) {
+        p_time_stat=layers[i]->get_time_stat();
+
+        p_time_stat->count=0;
+        p_time_stat->total=0;
+        p_time_stat->used=p_time_stat->start=p_time_stat->end=0;
+   }
+}
+
+void Classifier::dump_all_stat(vector<vector<perf_stat>*>& all_stat)
+{
+
+   struct new_perf_stat {
+        perf_stat stat;
+        int       idx;
+   };
+    
+   vector<new_perf_stat > layer_stat;
+   perf_stat * p_stat;
+
+   uint64_t total_time=0;
+
+   layer_stat.resize(all_stat[0]->size());
+
+   for(int i=0;i<all_stat.size();i++)
+   {
+      for(int j=0;j<layer_stat.size();j++)
+       {
+          p_stat=&layer_stat[j].stat;
+
+          p_stat->total+=(*all_stat[i])[j].total;
+          p_stat->count+=(*all_stat[i])[j].count;
+          total_time+=(*all_stat[i])[j].total;
+       }
+   }
+
+   total_time=total_time/all_stat.size();
+
+   std::cout<<std::endl<<"----------------------------------"<<std::endl;
+   std::cout<<"STATS for "<<all_stat.size()<<" reptitions: ..."<<std::endl;
+   std::cout<<"Total time: "<<total_time<<" per forward"<<std::endl;
+   std::cout<<"Each layer stats: ..."<<std::endl;
+
+
+   for(int i=layer_stat.size()-1;i>=0;i--)
+   {
+      p_stat=&layer_stat[i].stat;
+
+      layer_stat[i].idx=i;
+
+     std::cout<<"  "<<i<<": used time: "<<p_stat->total/all_stat.size();
+     std::cout<<" ratio: "<<((float)p_stat->total)/all_stat.size()/total_time*100;
+     std::cout<<" enter count: "<<p_stat->count/all_stat.size()<<std::endl;
+   }
+
+   std::cout<<std::endl;
+
+   std::cout<<"time cost top 10 layers are: ..."<<std::endl;
+
+   std::sort(layer_stat.begin(),layer_stat.end(),[](const new_perf_stat& a, const new_perf_stat& b)
+       {
+          if(a.stat.total>b.stat.total)
+            return true;
+          else
+            return false;
+       });
+
+   uint64_t  top_total_time=0;
+
+   for(int i=0; i<10; i++)
+   {
+      p_stat=&layer_stat[i].stat;
+
+     std::cout<<"  "<<layer_stat[i].idx<<": used time: "<<p_stat->total/all_stat.size();
+     std::cout<<" ratio: "<<((float)p_stat->total)/all_stat.size()/total_time*100;
+     std::cout<<" enter count: "<<p_stat->count/all_stat.size()<<std::endl;
+     top_total_time+=p_stat->total;
+   }
+
+   std::cout<<"Top cost layers occupied: "<<(float)top_total_time/all_stat.size()/total_time*100<<std::endl;
+
+   std::cout<<std::endl;
+}
+
+#endif
+
+void Classifier::dump_single_layer_io(int idx, Layer<float> * p_layer)
+{
+   const LayerParameter& layer_param=p_layer->layer_param();
+
+   std::cout<<std::endl<<"LAYER IDX: "<<idx<<" name: "<<layer_param.name();
+   std::cout<<" type: "<<layer_param.type()<<std::endl;
+
+   const vector<Blob<float>*> *p_bottom_vec=p_layer->saved_bottom;
+
+   for(int i=0;i<layer_param.bottom_size(); i++)
+   {
+      std::cout<<"bottom "<<layer_param.bottom(i)<<": ";
+
+      Blob<float> * p_blob=(*p_bottom_vec)[i];
+
+      for(int j=0;j<p_blob->num_axes();j++)
+      {
+          std::cout<<p_blob->shape(j)<<" ";
+      }
+      std::cout<<std::endl;
+   }
+
+   const vector<Blob<float>*> *p_top_vec=p_layer->saved_top;
+   for(int i=0;i<layer_param.top_size(); i++)
+   {
+      std::cout<<"top "<<layer_param.top(i)<<": ";
+      Blob<float> * p_blob=(*p_top_vec)[i];
+
+      for(int j=0;j<p_blob->num_axes();j++)
+      {
+          std::cout<<p_blob->shape(j)<<" ";
+      }
+      std::cout<<std::endl;
+   }
+}
+
+void Classifier::dump_single_layer_perf(int idx, Layer<float> * p_layer, uint64_t total_net_time)
+{
+   const LayerParameter& layer_param=p_layer->layer_param();
+   perf_stat * p_time_stat;
+
+   p_time_stat=p_layer->get_time_stat();
+
+   std::cout<<std::endl<<"LAYER IDX: "<<idx<<" name: "<<layer_param.name();
+   std::cout<<" type: "<<layer_param.type();
+   std::cout<<"  ratio: "<<(float)p_time_stat->total/total_net_time*100<<std::endl;
+
+
+   std::cout<<"time stat:  total: "<<p_time_stat->total<<" count: "<<p_time_stat->count;
+   if(p_time_stat->count)
+    {
+       std::cout<<" average: "<<((float)p_time_stat->total)/p_time_stat->count;
+    }
+
+   std::cout<<" start: "<<p_time_stat->start<<" end: "<<p_time_stat->end;
+   std::cout<<std::endl;
+
+
+} 
+
+void Classifier::dump_perf_stat(void)
+{
+   uint64_t total_net_time=0;
+
+   const vector<shared_ptr<Layer<float> > >& layers=net_->layers();
+
+   std::cout<<"Input/output shape for each layer ... total: "<<layers.size()<<std::endl;
+
+   for (int i = layers.size() - 1; i >= 0; --i) {
+     dump_single_layer_io(i,layers[i].get());
+   }
+
+
+   for (int i = layers.size() - 1; i >= 0; --i) {
+
+     perf_stat * p_time_stat;
+
+     p_time_stat=layers[i]->get_time_stat();
+
+     total_net_time+=p_time_stat->total;
+
+   }
+   
+   std::cout<<"Time for each layer ... sum of all layers is : ";
+   std::cout<<total_net_time<<std::endl;
+
+   for (int i = layers.size() - 1; i >= 0; --i) {
+
+     dump_single_layer_perf(i,layers[i].get(),total_net_time);
+   }
+
+}
+
+#endif
+
+#endif //USE_PROFILING
+
+/* Wrap the input layer of the network in separate cv::Mat objects
+ * (one per channel). This way we save one memcpy operation and we
+ * don't need to rely on cudaMemcpy2D. The last preprocessing
+ * operation will write the separate channels directly to the input
+ * layer. */
+void Classifier::WrapInputLayer(std::vector<cv::Mat>* input_channels) {
+  Blob<float>* input_layer = net_->input_blobs()[0];
+
+  int width = input_layer->width();
+  int height = input_layer->height();
+  float* input_data = input_layer->mutable_cpu_data();
+  for (int i = 0; i < input_layer->channels(); ++i) {
+    cv::Mat channel(height, width, CV_32FC1, input_data);
+    input_channels->push_back(channel);
+    input_data += width * height;
+  }
+}
+
+void Classifier::Preprocess(const cv::Mat& img,
+                            std::vector<cv::Mat>* input_channels) {
+  /* Convert the input image to the input image format of the network. */
+  cv::Mat sample;
+  if (img.channels() == 3 && num_channels_ == 1)
+    cv::cvtColor(img, sample, cv::COLOR_BGR2GRAY);
+  else if (img.channels() == 4 && num_channels_ == 1)
+    cv::cvtColor(img, sample, cv::COLOR_BGRA2GRAY);
+  else if (img.channels() == 4 && num_channels_ == 3)
+    cv::cvtColor(img, sample, cv::COLOR_BGRA2BGR);
+  else if (img.channels() == 1 && num_channels_ == 3)
+    cv::cvtColor(img, sample, cv::COLOR_GRAY2BGR);
+  else
+    sample = img;
+
+  cv::Mat sample_resized;
+  if (sample.size() != input_geometry_)
+    cv::resize(sample, sample_resized, input_geometry_);
+  else
+    sample_resized = sample;
+
+  cv::Mat sample_float;
+  if (num_channels_ == 3)
+    sample_resized.convertTo(sample_float, CV_32FC3);
+  else
+    sample_resized.convertTo(sample_float, CV_32FC1);
+
+  cv::Mat sample_normalized;
+  cv::subtract(sample_float, mean_, sample_normalized);
+
+  /* This operation will write the separate BGR planes directly to the
+   * input layer of the network because it is wrapped by the cv::Mat
+   * objects in input_channels. */
+  cv::split(sample_normalized, *input_channels);
+
+  CHECK(reinterpret_cast<float*>(input_channels->at(0).data)
+        == net_->input_blobs()[0]->cpu_data())
+    << "Input channels are not wrapping the input layer of the network.";
+}
+
+int main(int argc, char** argv) {
+  if (argc != 6) {
+    std::cerr << "Usage: " << argv[0]
+              << " deploy.prototxt network.caffemodel"
+              << " mean.binaryproto labels.txt img.jpg" << std::endl;
+    return 1;
+  }
+
+  ::google::InitGoogleLogging(argv[0]);
+
+  string model_file   = argv[1];
+  string trained_file = argv[2];
+  string mean_file    = argv[3];
+  string label_file   = argv[4];
+  Classifier classifier(model_file, trained_file, mean_file, label_file);
+
+  string file = argv[5];
+
+  std::cout << "---------- Prediction for "
+            << file << " ----------" << std::endl;
+
+  cv::Mat img = cv::imread(file, -1);
+  CHECK(!img.empty()) << "Unable to decode image " << file;
+  std::vector<Prediction> predictions = classifier.Classify(img);
+
+  /* Print the top N predictions. */
+  for (size_t i = 0; i < predictions.size(); ++i) {
+    Prediction p = predictions[i];
+    std::cout << std::fixed << std::setprecision(4) << p.second << " - \""
+              << p.first << "\"" << std::endl;
+  }
+}
+#else
+int main(int argc, char** argv) {
+  LOG(FATAL) << "This example requires OpenCV; compile with USE_OPENCV.";
+}
+#endif  // USE_OPENCV
diff --git a/examples/cpp_classification/classification_profiling_gpu.cpp b/examples/cpp_classification/classification_profiling_gpu.cpp
new file mode 100644
index 00000000..3c5e04ad
--- /dev/null
+++ b/examples/cpp_classification/classification_profiling_gpu.cpp
@@ -0,0 +1,546 @@
+#include <caffe/caffe.hpp>
+#ifdef USE_OPENCV
+#include <opencv2/core/core.hpp>
+#include <opencv2/highgui/highgui.hpp>
+#include <opencv2/imgproc/imgproc.hpp>
+#endif  // USE_OPENCV
+#include <algorithm>
+#include <iosfwd>
+#include <memory>
+#include <string>
+#include <utility>
+#include <vector>
+
+#ifdef USE_PROFILING
+
+#include <iostream>
+
+#include <time.h>
+
+#define REPEAT_TEST
+
+unsigned long get_cur_time(void)
+{
+   struct timespec tm;
+
+   clock_gettime(CLOCK_MONOTONIC_COARSE, &tm);
+
+   return (tm.tv_sec*1000+tm.tv_nsec/1000000);
+}
+
+#endif //USE_PROFILING
+
+#ifdef USE_OPENCV
+using namespace caffe;  // NOLINT(build/namespaces)
+using std::string;
+
+/* Pair (label, confidence) representing a prediction. */
+typedef std::pair<string, float> Prediction;
+
+class Classifier {
+ public:
+  Classifier(const string& model_file,
+             const string& trained_file,
+             const string& mean_file,
+             const string& label_file);
+
+  std::vector<Prediction> Classify(const cv::Mat& img, int N = 5);
+
+#ifdef USE_PROFILING
+
+#ifdef LAYER_PERF_STAT
+  void  dump_perf_stat(void);
+  void  dump_single_layer_io(int idx, Layer<float> * p_layer);
+  void  dump_single_layer_perf(int idx, Layer<float> * p_layer,uint64_t total_net_time);
+#ifdef REPEAT_TEST
+  void collect_layer_stat(vector<vector<perf_stat> * > & all_stat);
+  void dump_all_stat(vector <vector<perf_stat>*>& all_stat);
+  void reset_layer_stat();
+#endif
+#endif
+
+#endif //USE_PROFILING
+
+ private:
+  void SetMean(const string& mean_file);
+
+  std::vector<float> Predict(const cv::Mat& img);
+
+  void WrapInputLayer(std::vector<cv::Mat>* input_channels);
+
+  void Preprocess(const cv::Mat& img,
+                  std::vector<cv::Mat>* input_channels);
+
+ private:
+  shared_ptr<Net<float> > net_;
+  cv::Size input_geometry_;
+  int num_channels_;
+  cv::Mat mean_;
+  std::vector<string> labels_;
+};
+
+Classifier::Classifier(const string& model_file,
+                       const string& trained_file,
+                       const string& mean_file,
+                       const string& label_file) {
+//#ifdef CPU_ONLY
+//  Caffe::set_mode(Caffe::CPU);
+//#else
+  Caffe::set_mode(Caffe::GPU); //For ARM GPU (the code is in CPU_ONLY mode, just set caffe mode to GPU)
+//#endif
+
+  /* Load the network. */
+  net_.reset(new Net<float>(model_file, TEST));
+  net_->CopyTrainedLayersFrom(trained_file);
+
+  CHECK_EQ(net_->num_inputs(), 1) << "Network should have exactly one input.";
+  CHECK_EQ(net_->num_outputs(), 1) << "Network should have exactly one output.";
+
+  Blob<float>* input_layer = net_->input_blobs()[0];
+  num_channels_ = input_layer->channels();
+  CHECK(num_channels_ == 3 || num_channels_ == 1)
+    << "Input layer should have 1 or 3 channels.";
+  input_geometry_ = cv::Size(input_layer->width(), input_layer->height());
+
+  /* Load the binaryproto mean file. */
+  SetMean(mean_file);
+
+  /* Load labels. */
+  std::ifstream labels(label_file.c_str());
+  CHECK(labels) << "Unable to open labels file " << label_file;
+  string line;
+  while (std::getline(labels, line))
+    labels_.push_back(string(line));
+
+  Blob<float>* output_layer = net_->output_blobs()[0];
+  CHECK_EQ(labels_.size(), output_layer->channels())
+    << "Number of labels is different from the output layer dimension.";
+}
+
+static bool PairCompare(const std::pair<float, int>& lhs,
+                        const std::pair<float, int>& rhs) {
+  return lhs.first > rhs.first;
+}
+
+/* Return the indices of the top N values of vector v. */
+static std::vector<int> Argmax(const std::vector<float>& v, int N) {
+  std::vector<std::pair<float, int> > pairs;
+  for (size_t i = 0; i < v.size(); ++i)
+    pairs.push_back(std::make_pair(v[i], i));
+  std::partial_sort(pairs.begin(), pairs.begin() + N, pairs.end(), PairCompare);
+
+  std::vector<int> result;
+  for (int i = 0; i < N; ++i)
+    result.push_back(pairs[i].second);
+  return result;
+}
+
+/* Return the top N predictions. */
+std::vector<Prediction> Classifier::Classify(const cv::Mat& img, int N) {
+  std::vector<float> output = Predict(img);
+
+  N = std::min<int>(labels_.size(), N);
+  std::vector<int> maxN = Argmax(output, N);
+  std::vector<Prediction> predictions;
+  for (int i = 0; i < N; ++i) {
+    int idx = maxN[i];
+    predictions.push_back(std::make_pair(labels_[idx], output[idx]));
+  }
+
+  return predictions;
+}
+
+/* Load the mean file in binaryproto format. */
+void Classifier::SetMean(const string& mean_file) {
+  BlobProto blob_proto;
+  ReadProtoFromBinaryFileOrDie(mean_file.c_str(), &blob_proto);
+
+  /* Convert from BlobProto to Blob<float> */
+  Blob<float> mean_blob;
+  mean_blob.FromProto(blob_proto);
+  CHECK_EQ(mean_blob.channels(), num_channels_)
+    << "Number of channels of mean file doesn't match input layer.";
+
+  /* The format of the mean file is planar 32-bit float BGR or grayscale. */
+  std::vector<cv::Mat> channels;
+  float* data = mean_blob.mutable_cpu_data();
+  for (int i = 0; i < num_channels_; ++i) {
+    /* Extract an individual channel. */
+    cv::Mat channel(mean_blob.height(), mean_blob.width(), CV_32FC1, data);
+    channels.push_back(channel);
+    data += mean_blob.height() * mean_blob.width();
+  }
+
+  /* Merge the separate channels into a single image. */
+  cv::Mat mean;
+  cv::merge(channels, mean);
+
+  /* Compute the global mean pixel value and create a mean image
+   * filled with this value. */
+  cv::Scalar channel_mean = cv::mean(mean);
+  mean_ = cv::Mat(input_geometry_, mean.type(), channel_mean);
+}
+
+std::vector<float> Classifier::Predict(const cv::Mat& img) {
+  Blob<float>* input_layer = net_->input_blobs()[0];
+  input_layer->Reshape(1, num_channels_,
+                       input_geometry_.height, input_geometry_.width);
+  /* Forward dimension change to all layers. */
+  net_->Reshape();
+
+  std::vector<cv::Mat> input_channels;
+  WrapInputLayer(&input_channels);
+
+  Preprocess(img, &input_channels);
+
+#ifdef USE_PROFILING
+  unsigned long tstart=get_cur_time();
+#endif //USE_PROFILING
+
+  net_->Forward();
+
+#ifdef USE_PROFILING
+
+  unsigned long tend=get_cur_time();
+
+  std::cout<<"used time: "<<tend-tstart<<std::endl;
+
+#ifdef LAYER_PERF_STAT
+  dump_perf_stat(); 
+#ifdef REPEAT_TEST
+
+   reset_layer_stat();
+
+   vector<vector<perf_stat>* >  all_stat;
+   int rep_number=10;
+
+   for(int i=0;i<rep_number;i++)
+   {
+      net_->Forward();
+      collect_layer_stat(all_stat);
+      reset_layer_stat();
+   }
+
+   //dump stats
+   dump_all_stat(all_stat);
+
+   for(int i=0;i<all_stat.size();i++)
+         delete all_stat[i];
+   
+#endif //REPEAT_TEST
+#endif //LAYER_PERF_STAT
+#endif //USE_PROFILING
+
+  /* Copy the output layer to a std::vector */
+  Blob<float>* output_layer = net_->output_blobs()[0];
+  const float* begin = output_layer->cpu_data();
+  const float* end = begin + output_layer->channels();
+  return std::vector<float>(begin, end);
+}
+
+#ifdef USE_PROFILING
+
+#ifdef LAYER_PERF_STAT
+
+#ifdef REPEAT_TEST
+void Classifier::collect_layer_stat(vector<vector<perf_stat>*>& all_stat)
+{
+   vector<perf_stat > * p_stat;
+   perf_stat * p_time_stat;
+   const vector<shared_ptr<Layer<float> > >& layers=net_->layers();
+
+   
+   p_stat=new vector<perf_stat>;
+
+   for (int i =0;i< layers.size(); i++) {
+        p_time_stat=layers[i]->get_time_stat();
+        p_stat->push_back(*p_time_stat);
+
+   }
+
+   all_stat.push_back(p_stat);
+}
+
+void Classifier::reset_layer_stat(void)
+{
+   const vector<shared_ptr<Layer<float> > >& layers=net_->layers();
+   perf_stat * p_time_stat;
+
+   for (int i =0;i< layers.size(); i++) {
+        p_time_stat=layers[i]->get_time_stat();
+
+        p_time_stat->count=0;
+        p_time_stat->total=0;
+        p_time_stat->used=p_time_stat->start=p_time_stat->end=0;
+   }
+}
+
+void Classifier::dump_all_stat(vector<vector<perf_stat>*>& all_stat)
+{
+
+   struct new_perf_stat {
+        perf_stat stat;
+        int       idx;
+   };
+    
+   vector<new_perf_stat > layer_stat;
+   perf_stat * p_stat;
+
+   uint64_t total_time=0;
+
+   layer_stat.resize(all_stat[0]->size());
+
+   for(int i=0;i<all_stat.size();i++)
+   {
+      for(int j=0;j<layer_stat.size();j++)
+       {
+          p_stat=&layer_stat[j].stat;
+
+          p_stat->total+=(*all_stat[i])[j].total;
+          p_stat->count+=(*all_stat[i])[j].count;
+          total_time+=(*all_stat[i])[j].total;
+       }
+   }
+
+   total_time=total_time/all_stat.size();
+
+   std::cout<<std::endl<<"----------------------------------"<<std::endl;
+   std::cout<<"STATS for "<<all_stat.size()<<" reptitions: ..."<<std::endl;
+   std::cout<<"Total time: "<<total_time<<" per forward"<<std::endl;
+   std::cout<<"Each layer stats: ..."<<std::endl;
+
+
+   for(int i=layer_stat.size()-1;i>=0;i--)
+   {
+      p_stat=&layer_stat[i].stat;
+
+      layer_stat[i].idx=i;
+
+     std::cout<<"  "<<i<<": used time: "<<p_stat->total/all_stat.size();
+     std::cout<<" ratio: "<<((float)p_stat->total)/all_stat.size()/total_time*100;
+     std::cout<<" enter count: "<<p_stat->count/all_stat.size()<<std::endl;
+   }
+
+   std::cout<<std::endl;
+
+   std::cout<<"time cost top 10 layers are: ..."<<std::endl;
+
+   std::sort(layer_stat.begin(),layer_stat.end(),[](const new_perf_stat& a, const new_perf_stat& b)
+       {
+          if(a.stat.total>b.stat.total)
+            return true;
+          else
+            return false;
+       });
+
+   uint64_t  top_total_time=0;
+
+   for(int i=0; i<10; i++)
+   {
+      p_stat=&layer_stat[i].stat;
+
+     std::cout<<"  "<<layer_stat[i].idx<<": used time: "<<p_stat->total/all_stat.size();
+     std::cout<<" ratio: "<<((float)p_stat->total)/all_stat.size()/total_time*100;
+     std::cout<<" enter count: "<<p_stat->count/all_stat.size()<<std::endl;
+     top_total_time+=p_stat->total;
+   }
+
+   std::cout<<"Top cost layers occupied: "<<(float)top_total_time/all_stat.size()/total_time*100<<std::endl;
+
+   std::cout<<std::endl;
+}
+
+#endif
+
+void Classifier::dump_single_layer_io(int idx, Layer<float> * p_layer)
+{
+   const LayerParameter& layer_param=p_layer->layer_param();
+
+   std::cout<<std::endl<<"LAYER IDX: "<<idx<<" name: "<<layer_param.name();
+   std::cout<<" type: "<<layer_param.type()<<std::endl;
+
+   const vector<Blob<float>*> *p_bottom_vec=p_layer->saved_bottom;
+
+   for(int i=0;i<layer_param.bottom_size(); i++)
+   {
+      std::cout<<"bottom "<<layer_param.bottom(i)<<": ";
+
+      Blob<float> * p_blob=(*p_bottom_vec)[i];
+
+      for(int j=0;j<p_blob->num_axes();j++)
+      {
+          std::cout<<p_blob->shape(j)<<" ";
+      }
+      std::cout<<std::endl;
+   }
+
+   const vector<Blob<float>*> *p_top_vec=p_layer->saved_top;
+   for(int i=0;i<layer_param.top_size(); i++)
+   {
+      std::cout<<"top "<<layer_param.top(i)<<": ";
+      Blob<float> * p_blob=(*p_top_vec)[i];
+
+      for(int j=0;j<p_blob->num_axes();j++)
+      {
+          std::cout<<p_blob->shape(j)<<" ";
+      }
+      std::cout<<std::endl;
+   }
+}
+
+void Classifier::dump_single_layer_perf(int idx, Layer<float> * p_layer, uint64_t total_net_time)
+{
+   const LayerParameter& layer_param=p_layer->layer_param();
+   perf_stat * p_time_stat;
+
+   p_time_stat=p_layer->get_time_stat();
+
+   std::cout<<std::endl<<"LAYER IDX: "<<idx<<" name: "<<layer_param.name();
+   std::cout<<" type: "<<layer_param.type();
+   std::cout<<"  ratio: "<<(float)p_time_stat->total/total_net_time*100<<std::endl;
+
+
+   std::cout<<"time stat:  total: "<<p_time_stat->total<<" count: "<<p_time_stat->count;
+   if(p_time_stat->count)
+    {
+       std::cout<<" average: "<<((float)p_time_stat->total)/p_time_stat->count;
+    }
+
+   std::cout<<" start: "<<p_time_stat->start<<" end: "<<p_time_stat->end;
+   std::cout<<std::endl;
+
+
+} 
+
+void Classifier::dump_perf_stat(void)
+{
+   uint64_t total_net_time=0;
+
+   const vector<shared_ptr<Layer<float> > >& layers=net_->layers();
+
+   std::cout<<"Input/output shape for each layer ... total: "<<layers.size()<<std::endl;
+
+   for (int i = layers.size() - 1; i >= 0; --i) {
+     dump_single_layer_io(i,layers[i].get());
+   }
+
+
+   for (int i = layers.size() - 1; i >= 0; --i) {
+
+     perf_stat * p_time_stat;
+
+     p_time_stat=layers[i]->get_time_stat();
+
+     total_net_time+=p_time_stat->total;
+
+   }
+   
+   std::cout<<"Time for each layer ... sum of all layers is : ";
+   std::cout<<total_net_time<<std::endl;
+
+   for (int i = layers.size() - 1; i >= 0; --i) {
+
+     dump_single_layer_perf(i,layers[i].get(),total_net_time);
+   }
+
+}
+
+#endif
+
+#endif //USE_PROFILING
+
+/* Wrap the input layer of the network in separate cv::Mat objects
+ * (one per channel). This way we save one memcpy operation and we
+ * don't need to rely on cudaMemcpy2D. The last preprocessing
+ * operation will write the separate channels directly to the input
+ * layer. */
+void Classifier::WrapInputLayer(std::vector<cv::Mat>* input_channels) {
+  Blob<float>* input_layer = net_->input_blobs()[0];
+
+  int width = input_layer->width();
+  int height = input_layer->height();
+  float* input_data = input_layer->mutable_cpu_data();
+  for (int i = 0; i < input_layer->channels(); ++i) {
+    cv::Mat channel(height, width, CV_32FC1, input_data);
+    input_channels->push_back(channel);
+    input_data += width * height;
+  }
+}
+
+void Classifier::Preprocess(const cv::Mat& img,
+                            std::vector<cv::Mat>* input_channels) {
+  /* Convert the input image to the input image format of the network. */
+  cv::Mat sample;
+  if (img.channels() == 3 && num_channels_ == 1)
+    cv::cvtColor(img, sample, cv::COLOR_BGR2GRAY);
+  else if (img.channels() == 4 && num_channels_ == 1)
+    cv::cvtColor(img, sample, cv::COLOR_BGRA2GRAY);
+  else if (img.channels() == 4 && num_channels_ == 3)
+    cv::cvtColor(img, sample, cv::COLOR_BGRA2BGR);
+  else if (img.channels() == 1 && num_channels_ == 3)
+    cv::cvtColor(img, sample, cv::COLOR_GRAY2BGR);
+  else
+    sample = img;
+
+  cv::Mat sample_resized;
+  if (sample.size() != input_geometry_)
+    cv::resize(sample, sample_resized, input_geometry_);
+  else
+    sample_resized = sample;
+
+  cv::Mat sample_float;
+  if (num_channels_ == 3)
+    sample_resized.convertTo(sample_float, CV_32FC3);
+  else
+    sample_resized.convertTo(sample_float, CV_32FC1);
+
+  cv::Mat sample_normalized;
+  cv::subtract(sample_float, mean_, sample_normalized);
+
+  /* This operation will write the separate BGR planes directly to the
+   * input layer of the network because it is wrapped by the cv::Mat
+   * objects in input_channels. */
+  cv::split(sample_normalized, *input_channels);
+
+  CHECK(reinterpret_cast<float*>(input_channels->at(0).data)
+        == net_->input_blobs()[0]->cpu_data())
+    << "Input channels are not wrapping the input layer of the network.";
+}
+
+int main(int argc, char** argv) {
+  if (argc != 6) {
+    std::cerr << "Usage: " << argv[0]
+              << " deploy.prototxt network.caffemodel"
+              << " mean.binaryproto labels.txt img.jpg" << std::endl;
+    return 1;
+  }
+
+  ::google::InitGoogleLogging(argv[0]);
+
+  string model_file   = argv[1];
+  string trained_file = argv[2];
+  string mean_file    = argv[3];
+  string label_file   = argv[4];
+  Classifier classifier(model_file, trained_file, mean_file, label_file);
+
+  string file = argv[5];
+
+  std::cout << "---------- Prediction for "
+            << file << " ----------" << std::endl;
+
+  cv::Mat img = cv::imread(file, -1);
+  CHECK(!img.empty()) << "Unable to decode image " << file;
+  std::vector<Prediction> predictions = classifier.Classify(img);
+
+  /* Print the top N predictions. */
+  for (size_t i = 0; i < predictions.size(); ++i) {
+    Prediction p = predictions[i];
+    std::cout << std::fixed << std::setprecision(4) << p.second << " - \""
+              << p.first << "\"" << std::endl;
+  }
+}
+#else
+int main(int argc, char** argv) {
+  LOG(FATAL) << "This example requires OpenCV; compile with USE_OPENCV.";
+}
+#endif  // USE_OPENCV