What if I have (say) three BBs per segment, and one segment is flooded with a million recognized objects? Think of a bee detector, looking at an image with a swarm of bees in the center. Most of the bee BBs (sorry!) will B empty (sorry again!) , but the ones with the swarm will be overwhelmed. Would I detect only three B’s per BB?
I wrote a detailed blog post about how these bounding boxes work: https://machinethink.net/blog/object-detection/ – I hope it helps!