Pedestrian Attribute Recognition in Video Surveillance Scenarios Based on View-attribute Attention Localization

Pedestrian attribute recognition in surveillance scenarios is still a challenging task due to the inaccurate localization of specific attributes. In this paper, we propose novel view-attribute method based on attention (VALA), which utilizes view information guide process focus attributes and mechanism localize attribute-corresponding areas. Concretely, leveraged by prediction branch generate four weights that represent confidences for from different views. View are then delivered back compose view-attributes, will participate supervise deep feature extraction. order explore spatial location view-attribute, regional introduced aggregate encode inter-channel dependencies feature. Subsequently, fine attentive attribute-specific region localized, locations gained attention. The final outcome obtained combining with weights. Experiments three wide datasets (richly annotated pedestrian (RAP), v2 (RAPv2), PA-100K) demonstrate effectiveness our approach compared state-of-the-art methods.