Reasoning Graph Networks for Kinship Verification: from Star-shaped to Hierarchical

In this paper, we investigate the problem of facial kinship verification bylearning hierarchical reasoning graph networks. Conventional methods usuallyfocus on learning discriminative features for each facial image of a pairedsample and neglect how to fuse the obtained two facial image features andreason about the relations between them. To address this, we propose aStar-shaped Reasoning Graph Network (S-RGN). Our S-RGN first constructs astar-shaped graph where each surrounding node encodes the information ofcomparisons in a feature dimension and the central node is employed as thebridge for the interaction of surrounding nodes. Then we perform relationalreasoning on this star graph with iterative message passing. The proposed S-RGNuses only one central node to analyze and process information from allsurrounding nodes, which limits its reasoning capacity. We further develop aHierarchical Reasoning Graph Network (H-RGN) to exploit more powerful andflexible capacity. More specifically, our H-RGN introduces a set of latentreasoning nodes and constructs a hierarchical graph with them. Then bottom-upcomparative information abstraction and top-down comprehensive signalpropagation are iteratively performed on the hierarchical graph to update thenode features. Extensive experimental results on four widely used kinshipdatabases show that the proposed methods achieve very competitive results.