two classifiers are better than one

we can fuse detection boxes from different models ,
eg. 2 different yolo detectors, or even SSD against a flipped version
(test time data augmentation)

wbf nano yolo4

also errors happen (joined 2 different persons):

in opencv/c++ it might look like this:

https://github.com/ZFTurbo/Weighted-Boxes-Fusion