2020/04/16 by Siddharth Tourani, Tourani, Siddharth, Alexander Shekhovtsov +5 · 1 citation
Computer Science · Engineering · #Advanced Image and Video Retrieval Techniques #Advanced Neural Network Applications #Computer Vision and Pattern Recognition (cs.CV) #FOS: Computer and information sciences #Machine Learning (cs.LG) #Machine Learning (stat.ML) #Robotics and Sensor-Based Localization
paper · pdf · doi:10.48550/arxiv.2004.08227
openalex publication_date 2020/04/16 · openalex created_date 2022/07/26 · openalex updated_date 2026/07/28
Dense, discrete Graphical Models with pairwise potentials are a powerful\nclass of models which are employed in state-of-the-art computer vision and\nbio-imaging applications. This work introduces a new MAP-solver, based on the\npopular Dual Block-Coordinate Ascent principle. Surprisingly, by making a small\nchange to the low-performing solver, the Max Product Linear Programming (MPLP)\nalgorithm, we derive the new solver MPLP++ that significantly outperforms all\nexisting solvers by a large margin, including the state-of-the-art solver\nTree-Reweighted Sequential (TRWS) message-passing algorithm. Additionally, our\nsolver is highly parallel, in contrast to TRWS, which gives a further boost in\nperformance with the proposed GPU and multi-thread CPU implementations. We\nverify the superiority of our algorithm on dense problems from publicly\navailable benchmarks, as well, as a new benchmark for 6D Object Pose\nestimation. We also provide an ablation study with respect to graph density.\n