Efficient Semidefinite Branch-and-Cut for MAP-MRF Inference

We propose a new Branch-and-Cut (BC) method for solving general MAP-MRF inference problems. The core of our method is a very efficient bounding procedure, which combines scalable semidefinite programming (SDP) and a cutting-plane method for seeking violated constraints. We analyze the performance of the proposed method under different settings, and demonstrate that our method outperforms state-of-the-art, especially when connectivity is high or the relative magnitudes of the unary costs are low. Experiments show that our method achieves better approximation than the state-of-the-art methods within a variety of time budgets on difficult MAP-MRF inference problems.