detect.cpp

#include "anchor_generator.h"
#include "detect.h"
#include "tools.h"
#include<iostream>


using namespace std;
using namespace cv;
using namespace caffe;
uint64_t current_timestamp() {
    struct timeval te; 
    gettimeofday(&te, NULL); // get current time
    return te.tv_sec*1000LL + te.tv_usec/1000; // caculate milliseconds
}

/*void printMat(const cv::Mat &image)
{
    uint8_t  *myData = image.data;
    int width = image.cols;
    int height = image.rows;
    int _stride = image.step;//in case cols != strides
    for(int i = 0; i < height; i++)
    {
	for(int j = 0; j < width; j++)
	{
	    uint8_t  val = myData[ i * _stride + j];
	    cout << val;

	    //do whatever you want with your value
	}
    }
    cout << endl;
}*/



Detector::Detector(const string& model_file,
                   const string& weights_file,                  
                   const float confidence,
                   const float nms,                  
                   const string& gpu_mode) 
{
  if(gpu_mode == "gpu") 
  {
     #ifdef CPU_ONLY
        Caffe::set_mode(Caffe::CPU);
    #else
        Caffe::set_mode(Caffe::GPU);
        Caffe::SetDevice(0);
    #endif // CPU_ONLY  
     
  }
  else
  {
     Caffe::set_mode(Caffe::CPU);
  }
  /* Load the network. */
  net_.reset(new Net<float>(model_file, TEST));
  net_->CopyTrainedLayersFrom(weights_file);

  CHECK_EQ(net_->num_inputs(), 1) << "Network should have exactly one input.";
  Blob<float>* input_layer = net_->input_blobs()[0];
  num_channels_ = input_layer->channels();
  confidence_threshold =confidence;
  nms_threshold = nms;
  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());
}

void Detector::preprocess(const cv::Mat& img, std::vector<cv::Mat>* input_channels)
{

    cv::Mat sample, sample_resized, sample_float;
    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;
    //cout<<"img1"<<img.rows<<endl;
    //cv::cvtColor(img, sample, cv::COLOR_BGR2RGB);
    if(sample.size()!=input_geometry_)
    {
        //cout<<"need to be resized"<<endl;
        ratio_w = float(img.cols)/float(input_geometry_.width);
        ratio_h = float(img.rows)/float(input_geometry_.height);
        //cout<<"ratio_wratio_w"<<ratio_w  <<"ratio_hratio_h "<<ratio_h<<endl;
        cv::resize(sample, sample_resized, input_geometry_);
        //cout<<"img2"<<img.rows<<endl;
    }
    else
        sample_resized=sample;
    if (num_channels_ == 3)
        sample_resized.convertTo(sample_float, CV_32FC3);
    else
        sample_resized.convertTo(sample_float, CV_32FC1); 
    //cout<<"img3"<<img.rows<<endl;
    cv::split(sample_float, *input_channels);
}

void Detector::wrapInputLayer(std::shared_ptr<caffe::Net<float> > net_, std::vector<cv::Mat>* input_channels)
{
    caffe::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;
    }
}

std::vector<Anchor> Detector:: Detect(cv::Mat& img)
{
    //cout << "input.rows :"<<input.rows << "   "<< "fd_h_ :"<<fd_h_<<endl;
    //cout<<"input.cols :"<<input.cols <<"   "<<"fd_w_ :" << fd_w_<<endl;
    //cout<< "input.channels():"<<input.channels()<<"  "<<"fd_c_: "<<fd_c_<<endl;
    //assert(input.rows == fd_h_ && input.cols == fd_w_ && input.channels() == fd_c_);
    std::vector<cv::Mat> input_channels;
   
    wrapInputLayer(net_, &input_channels);   
    preprocess(img, &input_channels);
    net_->Forward();

    //extern std::vector<int> _feat_stride_fpn;
    //extern std::map<int, AnchorCfg> anchor_cfg;
    std::vector<AnchorGenerator> ac(_feat_stride_fpn.size());  //_feat_stride_fpn.size()=3  初始化容器容量为3
    for (int i = 0; i < _feat_stride_fpn.size(); ++i) {
        int stride = _feat_stride_fpn[i];
        ac[i].Init(stride, anchor_cfg[stride], false);
    }

    std::vector<Anchor> proposals;
    proposals.clear();


    for (int i = 0; i < _feat_stride_fpn.size(); ++i)
    {
        char clsname[100]; sprintf(clsname, "face_rpn_cls_prob_reshape_stride%d", _feat_stride_fpn[i]);
        char regname[100]; sprintf(regname, "face_rpn_bbox_pred_stride%d", _feat_stride_fpn[i]);
        char ptsname[100]; sprintf(ptsname, "face_rpn_landmark_pred_stride%d", _feat_stride_fpn[i]);
        const caffe::Blob<float>* clsBlob = net_->blob_by_name(clsname).get();
        const caffe::Blob<float>* regBlob = net_->blob_by_name(regname).get();
        const caffe::Blob<float>* ptsBlob = net_->blob_by_name(ptsname).get();
        ac[i].FilterAnchor(clsBlob, regBlob, ptsBlob, proposals,ratio_w,ratio_h,confidence_threshold);
        //printf("stride %d, res size %d\n", _feat_stride_fpn[i], proposals.size());


        /*for (int r = 0; r < proposals.size(); ++r) 
        {
            proposals[r].print();
        }*/

    }
    // nms
    std::vector<Anchor> result;
    nms_cpu(proposals, nms_threshold, result);

    //printf("final result %d\n", result.size());
    //result[0].print();
    return result;
}