Learning hybrid ranking representation for person re-identification

作者：

Highlights：

• We propose to jointly learn ranking context cues and appearance features to exploit discriminative feature representations for person re-id.

• We design a novel two-stream architecture to learn a hybrid ranking representation for more effective person re-id.

• Our method achieves superior performance compared with the state-of-the-art alternative methods on four large-scale person re-id benchmarks.

摘要

•We propose to jointly learn ranking context cues and appearance features to exploit discriminative feature representations for person re-id.•We design a novel two-stream architecture to learn a hybrid ranking representation for more effective person re-id.•Our method achieves superior performance compared with the state-of-the-art alternative methods on four large-scale person re-id benchmarks.

论文关键词：Person re-identification,Ranking representation,Ranking ensemble

论文评审过程：Received 26 August 2019, Revised 1 March 2021, Accepted 8 August 2021, Available online 9 August 2021, Version of Record 16 August 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.108239