Tie all bounding boxes to relevant images

[d33bs]

At the moment I believe we use one set of bounding box coordinates for all images. We should evolve this to use bounding box coordinates based on the compartment when available for a more truthful view of the objects.