Pedestrian Attribute Recognition Based on Association Rules

Over the past few years, deep learning has achieved impressive performance, and pedestrian attribute recognition has also been extensively widely studied. Pedestrian attribute recognition aims to predict a set of attributes from a predefined attributes list to describe the characteristics of the person. However, there are many different levels of attributes in the predefined attribute list, especially some high-level semantic information, so how to exploit the relationship between these attributes is an important challenge. We propose a flexible Association Rules Module(ARM), which can use association rules to express the relationship between attributes. Moreover, this module can work on different baselines. Extensive experiments show that the proposed method achieves excellent performance on two datasets and four baselines.