Neural network architecture to segment large objects from background in images

Question

I\'m currently working on an instance segmentation task where the goal is to retrieve the bounding rectangle of the top element of a stack of items displayed in a gray image.