HOME: 3D Human–Object Mesh Topology-Enhanced Interaction Recognition in Images

Human–object interaction (HOI) recognition is a very challenging task due to the ambiguity brought by occlusions, viewpoints, and poses. Because of limited information in image domain, extracting 3D features point cloud has been an important means improve performance HOI. However, neglect topological adjacent points at low level, deep topology relation between human object high level. In this paper, we present human–object mesh enhanced method (HOME) for HOI images. method, (HOM) built integrating reconstructed from images firstly. Therefore, under assumption that comes macroscopic pattern constructed spatial position microscopic human–object, HOM inputted into MeshCNN extract effective edge edge-based convolution bottom up, as encode invariance relationship. At last, cues are fused with visual enhance greatly. experiment, results have achieved improvement about 4.3% mean average precision (mAP) Rare cases HICO-DET dataset, which verifies effectiveness proposed method.