What makes an object an object? On a practical level, for computer analysis, what we have available is manual or “ground truth” stereotypes but what is better is connected pixels.
It’s part of the problem of “Salient Object Segmentation”.
ConnNet: A Long-Range Relation-Aware Pixel-Connectivity Network for Salient Segmentation
Michael Kampffmeyer, Nanqing Dong, Xiaodan Liang, Yujia Zhang, Eric P. Xing