AVPL: Augmented visual perception learning for person Re-identification and beyond
作者：
Highlights：
• Based on the perceptual characteristics of observation and the two-stream hypothesis, we propose a novel AVPL method for Person Re-identification, which considers two visual streams in HVS and explicitly models both of them.
• We propose an AVP module as the implementation of AVPL method in CNN-based network, which develops a bionics algorithm of v-d visual stream and enhances the representations in salient regions.
• We propose a novel BA method which provides a new perspective and approach for attention.
• To confirm the superiority of our method over a wide range of state-of-the-art RE-ID models, we extensively conduct our method on four large RE-ID benchmarks and other two recognition tasks.
摘要
•Based on the perceptual characteristics of observation and the two-stream hypothesis, we propose a novel AVPL method for Person Re-identification, which considers two visual streams in HVS and explicitly models both of them.•We propose an AVP module as the implementation of AVPL method in CNN-based network, which develops a bionics algorithm of v-d visual stream and enhances the representations in salient regions.•We propose a novel BA method which provides a new perspective and approach for attention.•To confirm the superiority of our method over a wide range of state-of-the-art RE-ID models, we extensively conduct our method on four large RE-ID benchmarks and other two recognition tasks.
论文关键词：Person Re-identification,Augmented visual perception learning,Batch attention,Two-stream hypothesis
论文评审过程：Received 9 November 2020, Revised 30 March 2022, Accepted 23 April 2022, Available online 26 April 2022, Version of Record 2 May 2022.
论文官网地址：https://doi.org/10.1016/j.patcog.2022.108736
