Test code for ObjectnessBING

bingtest.cpp

#include <opencv2/opencv.hpp>
#include <opencv2/saliency.hpp>

using namespace cv;
using namespace std;
using namespace saliency;

bool myfunction (int i, int j) { return (i<j); }

int main (int argc, const char * argv[]) {

  if (argc < 3) {
    printf("Usage: bingtest <image_file> <training_path>\n");
    printf("  <training_path> should point to opencv_contrib/modules/saliency/samples/ObjectnessTrainedModel or the directory where your own trained models are stored.\n");
    exit(-1);
  }

  Mat image = imread(argv[1]);
  string training_path = argv[2];
  imshow("original", image);

  Ptr<ObjectnessBING> objectnessBING = makePtr<ObjectnessBING>();
  objectnessBING->setTrainingPath(training_path);
  objectnessBING->setBBResDir(training_path + "/Results");

  int a = rand()*100;
  vector<Vec4i> objectnessBoundingBox;
  if (objectnessBING->computeSaliency(image, objectnessBoundingBox) ) {
    vector<float> values = objectnessBING->getobjectnessValues();

    printf("detected candidates: %d\n", objectnessBoundingBox.size());
    printf("scores: %d\n", values.size());

    // The result are sorted by objectness. We uonly use the first 20 boxes here.
    for (int i = 0; i < 20; i++) {
      Mat clone = image.clone();
      Vec4i bb = objectnessBoundingBox[i];
      printf("index=%d, value=%f\n", i, values[i]);
      rectangle(clone, Point(bb[0], bb[1]), Point(bb[2], bb[3]), Scalar(0, 0, 255), 4);

      char label[256];
      sprintf(label, "#%d", i+1);
      putText(clone, label, Point(bb[0], bb[1]+30), FONT_HERSHEY_SIMPLEX, 1, Scalar(0, 0, 255), 3);

      char filename[256];
      sprintf(filename, "bing_%05d.jpg", i);
      imwrite(filename, clone);
    }
  }
  return 0;
}