Unified graph-based method for instance separation from foreground-background segmentation

Instance segmentation problem is important for many computer vision applications. Recent state-of-the-art approaches use neural networks to predict per pixel foreground/background segmentation as well as other auxiliary data with subsequent methods for instance separation. All of these instance separation algorithms are very specific and require specific auxiliary data. Thus, we propose a unified graph-based method that can work with only foreground/background segmentation as well as with other auxiliary data structures.