Hello. I am testing with the yolo3 algorithm implemented in opencv the python version. I noticed however differences in results compared to the original pjreddie version of the algorithm. Using the same .cfg and .weights files. The opencv version cuts all results with a confidence threshold lower than 0.2. The getUnconnectedOutLayers function cuts all results below 0.2 of confidence to 0. Is it a bug? Do you know how to solve?
Thank you