Universe Points Representation Learning for Partial Multi-Graph Matching

Many challenges from natural world can be formulated as a graph matching problem. Previous deep learning-based methods mainly consider full two-graph setting. In this work, we study the more general partial problem with multi-graph cycle consistency guarantees. Building on recent progress in learning graphs, propose novel data-driven method (URL) for matching, which uses an object-to-universe formulation and learns latent representations of abstract universe points. The proposed approach advances state art semantic keypoint problem, evaluated Pascal VOC, CUB, Willow datasets. Moreover, set controlled experiments synthetic dataset demonstrates scalability our to graphs large number nodes its robustness high partiality.