Hi Jonathan,

This is a great post and I find it to be very useful!

You say the class confidence score=box confidence score x class conditional probability. However, I can imagine cases where a given bounding box sits on top of cells that have mixed predictions, i.e. one of the cells captures the tire of the bicycle in the background, and another cell captures the dog’s face.

In this case, does YOLO sum over the class conditional probabilities for all cells to determine what the most relevant object is in that bounding box?

Thanks for your time.