Ding Yiwei ; Deng Wenjin ; Zheng Yinglin ; Liu Pengfei ; Wang Meihong ; Cheng Xuan ; Bao Jianmin ; Chen Dong ; Zeng Ming

Abstract

In this paper, we present the Intra- and Inter-Human Relation Networks(I^2R-Net) for Multi-Person Pose Estimation. It involves two basic modules.First, the Intra-Human Relation Module operates on a single person and aims tocapture Intra-Human dependencies. Second, the Inter-Human Relation Moduleconsiders the relation between multiple instances and focuses on capturingInter-Human interactions. The Inter-Human Relation Module can be designed verylightweight by reducing the resolution of feature map, yet learn usefulrelation information to significantly boost the performance of the Intra-HumanRelation Module. Even without bells and whistles, our method can compete oroutperform current competition winners. We conduct extensive experiments onCOCO, CrowdPose, and OCHuman datasets. The results demonstrate that theproposed model surpasses all the state-of-the-art methods. Concretely, theproposed method achieves 77.4% AP on CrowPose dataset and 67.8% AP on OCHumandataset respectively, outperforming existing methods by a large margin.Additionally, the ablation study and visualization analysis also prove theeffectiveness of our model.