Closed liyuanyaun closed 6 years ago
score_cls
is the raw score before normalization (sigmoid). prob_cls
is probability score after normalization. Using either prob_cls or score_cls for NMS should not make a difference, because the relative order wasn't affected.
in tiny_face_detector.m, there are three output in your model:score_reg,prob_cls and score_cls. score_reg is to refine the detection windows. and while prob_cls > prob_thresh, the detection could be keep. and score_cls is used for nms. but what's the meaning of prob_cls and score_cls? why not prob_cls for nms? and i find the value of score_cls is more than 1. what is the score_cls represent? could anyone answer? thank you very much @peiyunh