cv::saliency::objectnecnessBING::getobjectnessValues() output

[vahidbas]

objectnecnessBING::getobjectnessValues() method should return saliency values for boxes return by algorithm. However, the values look like a sequence of integers which are calculated not inside the algorithm but in ObjectnessBING::ValStructVec<VT, ST>::append method (line 488, objectnessBING.cpp). I guess this method substitutes the values calculated by algo with these integers. I could not understand if its intentional or a bug!

[ramanpreet9]

I would like to know more about this as well. It should be returning values in decreasing order (i.e. bounding boxes with higher confidence should be ranked higher) but it returns value sin increasing order instead and looks like a bunch of integers increasing in value by 1-5.

[ignaciohrdz]

10 years later, I'm having the same issue. Has this been solved? Here is some basic Python code to test it. I got the models from here.

import cv2

bing_path = "..."
img_path = "..."

saliency_model = cv2.saliency.ObjectnessBING.create()
saliency_model.setTrainingPath(bing_path)

x = cv2.imread(img_path)
success, boxes = saliency_model.computeSaliency(x)

# The returned scores are in ascending order and they don't really look like scores
objectness = saliency_model.getobjectnessValues()