An Attention-Based Deep Learning Model for Pedestrian Attribute Recognition
Weiying Huang, Wenhua Li, An‐Min Zou · 2021 China Automation Congress (CAC) · 2021
Pedestrian attribute recognition, which can be applied in many scenes, has become one of the most popular research fields of computer vision, which can help person re-recognition and pedestrian retrieval. The paper proposes an neural network model with attentional mechanism for the pedestrian attribute recognition. The model consists of two modules: i) an attentional mechanism module to highlight features that are positive to the attribute recognition, which introduces a new weight learning method so that the module adjust the weight of features obtained in the training; ii) an attribute location module adaptively to locate attentional regions from the feature maps, which can localize the attentional region corresponding to individual attribute through the spatial transform module. The experiments achieve mean accuracy of 93.32% and 91.5% on Market1501 dataset and DukeMTMC-reID dataset respectively. The extensive experimental results illustrate the superiority and effectiveness of the proposed method.