Operating System / Platform => Ubuntu 16.04 64-bit
Compiler => gcc 5.4
Detailed description
The public function bool ObjectnessBING::computeSaliencyImpl(InputArray image, OutputArray objectnessBoundingBox) calls the private function void ObjectnessBING::getObjBndBoxesForSingleImage(Mat img, ValStructVec<float, Vec4i> &finalBoxes, int numDetPerSize) which requires a CV_8UC3 image and then proceeds to run the algorithm three times! on BGR input, HSV input and gray input, before printfing timing results (which are nowhere remotely close to 300Hz, more like 6Hz) and writing proposals to disk. Why? Presumably the user wants to choose between BGR, HSV, and gray, not printf timing to the console and not save results to disk.
The paper mentions the color spaces just once and the differences between running on one or all three are negligible.
My suggestions:
allow the user to select between BGR, HSV, and gray (or all 3, called "BING-diversified", see paper)
System information (version)
Detailed description
The public function
bool ObjectnessBING::computeSaliencyImpl(InputArray image, OutputArray objectnessBoundingBox)
calls the private functionvoid ObjectnessBING::getObjBndBoxesForSingleImage(Mat img, ValStructVec<float, Vec4i> &finalBoxes, int numDetPerSize)
which requires aCV_8UC3
image and then proceeds to run the algorithm three times! on BGR input, HSV input and gray input, beforeprintf
ing timing results (which are nowhere remotely close to 300Hz, more like 6Hz) and writing proposals to disk. Why? Presumably the user wants to choose between BGR, HSV, and gray, notprintf
timing to the console and not save results to disk.The paper mentions the color spaces just once and the differences between running on one or all three are negligible.
My suggestions:
Steps to reproduce
Examine the source code and run the sample.
Issue submission checklist