RESEARCH OF HUMAN ATTRIBUTES FOR THE TASK OF RE-IDENTIFICATION

Abstract

The paper investigates the problem of recognizing human attributes to improve re-identification. The analysis of re-identification methods and methods of recognition of human attributes is carried out. The data sets of human attributes are considered: Market-1501, DukeMTMC-reID, PETA, RAP-v1, RAP-v2 and PA-100k. Problems related to the formation and sets of human attributes and their use are analyzed. Recommendations for the selection and use of human attribute sets are provided. The paper also provides an overview of the metrics for evaluating methods for recognizing human attributes.