Solving Minimum Cost Lifted Multicut Problems by Node Agglomeration

Abstract

Despite its complexity, the minimum cost lifted multicut problem has found a wide range of applications in recent years, such as image and mesh decomposition or multiple object tracking. Its solutions are decompositions of a graph into an optimal number of segments which are optimized w.r.t. a cost function defined on a superset of the edge set. While the currently available solvers for this problem provide high quality solutions in terms of the task to be solved, they can have long computation times for more difficult problem instances. Here, we propose two variants of a heuristic solver (primal feasible heuristic), which greedily generate solutions within a bounded amount of time. Evaluations on image and mesh segmentation benchmarks show the high quality of these solutions.