Instead of splitting the space within the image equally each time, i wonder if you can split the image in half by the number of objects within the space using some type of a heuristic. For example, what if almost all objects were in section 4? Also, would using a heuristic be computationally efficient?
This is similar to the mipmap structure which means it's space efficient!
I wonder why we only split either the left or right node, rather than split both? For example, in this slide, does it mean that the number of objects in 1 is almost same with the sum of all object from 2 to 5?