Adaptive Hypergraph Neural Network for Multi-Person Pose Estimation

This paper proposes a novel two-stage hypergraph-based framework, dubbed ADaptive Hypergraph Neural Network (AD-HNN) to estimate multiple human poses from single image, with keypoint localization network and an Adaptive-Pose (AP-HNN) added onto the former network. For providing better guided representations of AP-HNN, we employ Semantic Interaction Convolution (SIC) module within initial acquire more explicit predictions. Build upon this, design adaptive hypergraph represent body for capturing high-order semantic relations among different joints. Notably, it can adaptively adjust between joints seek most reasonable structure variable benefit localization. These two stages are combined be trained in end-to-end fashion. Unlike traditional Graph Convolutional Networks (GCNs) that based on fixed tree structure, AP-HNN deal ambiguity pose estimation. Experimental results demonstrate AD-HNN achieves state-of-the-art performance both MS-COCO, MPII CrowdPose datasets.