Fast Fusion Moves for Multi-model Estimation

摘要: We develop a fast, effective algorithm for minimizing well-known objective function robust multi-model estimation. Our work introduces combinatorial step belonging to family of powerful move-making methods like α-expansion and fusion. also show that our subproblem can be quickly transformed into comparatively small instance minimum-weighted vertex-cover. In practice, these vertex-cover subproblems are almost always bipartite solved exactly by specialized network flow algorithms. Experiments indicate approach achieves the robustness affinity propagation, whilst providing speed fast greedy heuristics.