Currently, when you use cv::dnn::DetectionModel::detect, the non-max suppression does not suppress boxes if they are of different labels.

This can be bypassed by doing NMS manually, but I think it would be nice if there was some additional boolean to the function which would allow you to suppress boxes of different labels too

Reason: Sometimes, YoloV4 detects an object from 2 different points, and on one point it detects it as label X while on the other it detects it as label Y, resulting in 2 very similar boxes (ie a big overlap) for the same object.

As for which label and box should be chosen, I think the confidence value could be used to 'rank' the boxes, and the highest ranking one would remain.