Panoramic Human Activity Recognition

To obtain a more comprehensive activity understanding for crowded scene, in this paper, we propose new problem of panoramic human recognition (PAR), which aims to simultaneously achieve the individual actions, social group activities, and global activities. This is challenging yet practical real-world applications. track problem, develop novel hierarchical graph neural network progressively represent model multi-granular activities mutual relations crowd people. We further build benchmark evaluate proposed method other related methods. Experimental results verify rationality PAR effectiveness our usefulness benchmark. have released source code public promoting study on problem.