Hybrid attention network based on progressive embedding scale-context for crowd counting

The existing crowd counting methods usually adopt attention mechanisms to tackle background noise, or apply multilevel features multiscale context fusion scale variation. However, these approaches deal with two problems separately. In this paper, we propose a hybrid network (HAN) by employing progressive embedding scale-context (PES) information, which enables the simultaneously suppress noise and adapt head We build mechanism through parallel spatial channel modules, makes focus more on human area reduce interference of objects. addition, embed certain along dimensions alleviate errors caused variation perspective scale. Finally, learning strategy cascading multiple modules different contexts, can gradually integrate information into current feature map from global local. Ablation experiments show that architecture learn noise. Extensive demonstrate HANet obtains state-of-the-art performance five mainstream datasets.