Learning discriminative region representation for person retrieval

作者：

Highlights：

• We propose HRS, a weakly-supervised region segmentation method to predict informative region for improved person retrieval.

• HRS coherently incorporates global appearance information and local granularity cue in an end-to-end learning.

• Extensive experimental results on three benchmark datasets show HRS achieves new state-of-the-arts for person retrieval.

摘要

•We propose HRS, a weakly-supervised region segmentation method to predict informative region for improved person retrieval.•HRS coherently incorporates global appearance information and local granularity cue in an end-to-end learning.•Extensive experimental results on three benchmark datasets show HRS achieves new state-of-the-arts for person retrieval.

论文关键词：Person retrieval,Region representation

论文评审过程：Received 25 April 2021, Revised 25 July 2021, Accepted 6 August 2021, Available online 8 August 2021, Version of Record 13 August 2021.

论文官网地址：https://doi.org/10.1016/j.patcog.2021.108229