Hello everyone,
I am very excited about the ObjectnessBING algorithm. I have a data set with many roundly shaped objects. I expect the ObjectnessBING algorithm to perform better if the weight model is trained on my data set. Is this currently supported? I only found the ObjectnessBING::setTrainingPath function and none to train the my own model. How were the models in [1] generated? Is the structure of the *.yml.gz model files somewhere documented? In the BING paper a window of 500x10 pixels has a lower probability to be choosen then a 100x100 window. Is it possible to modify these probabilities?
[1] https://github.com/Itseez/opencv_contrib/tree/master/modules/saliency/samples/ObjectnessTrainedModel
