A Distributed Optimization Approach to Consistent Multiway Matching

Abstract

Multiway matching refers to the problem of establishing correspondences among a set of images from noisy pairwise correspondences, typically by exploiting cycle consistency. Existing approaches for multiway matching address the problem in a centralized setting. In this work, we propose a novel distributed optimization approach to multiway matching based on distributed projected gradient descent with constant step size. We rigorously analyze the convergence properties of our algorithm, specifically the range of the step size that guarantees convergence to a stationary point. We provide experimental evidence supporting that the proposed approach has performance comparable with the state of the art centralized approaches.