Rotation Synchronization via Deep Matrix Factorization
This addresses a key problem in computer vision (structure from motion and SLAM) with an incremental unsupervised approach.
The paper tackles the rotation synchronization problem by formulating it as an unsupervised deep matrix factorization task, achieving comparable accuracy to supervised competitors under weaker assumptions.
In this paper we address the rotation synchronization problem, where the objective is to recover absolute rotations starting from pairwise ones, where the unknowns and the measures are represented as nodes and edges of a graph, respectively. This problem is an essential task for structure from motion and simultaneous localization and mapping. We focus on the formulation of synchronization via neural networks, which has only recently begun to be explored in the literature. Inspired by deep matrix completion, we express rotation synchronization in terms of matrix factorization with a deep neural network. Our formulation exhibits implicit regularization properties and, more importantly, is unsupervised, whereas previous deep approaches are supervised. Our experiments show that we achieve comparable accuracy to the closest competitors in most scenes, while working under weaker assumptions.