"深度学习在行人重识别中的调查与展望"

"Deep Learning for Person Re-identification: A Survey and Outlook" is an article that discusses the application of deep learning in person re-identification. Person re-identification aims to retrieve interested individuals across multiple non-overlapping camera devices. With advancements in deep neural networks and the increasing demand in intelligent surveillance, person re-identification has gained significant interest in the field of computer vision. By analyzing the components involved in the development of a person re-identification system, the authors divide person re-identification into two categories: open-world and closed-world. Closed-world, which has been extensively researched, is applicable to various assumed scenarios and has achieved promising success using deep learning techniques on a number of datasets. The article starts with a deep analysis and comprehensive overview of closed-world person re-identification from three different aspects: deep representation learning, deep metric learning, and ranking optimization. As performance within the closed-world saturates, research in person re-identification has recently shifted towards the open-world, which poses more challenges and closely relates to real-world applications. The authors outline the open-world person re-identification from five different aspects. By analyzing the strengths of existing methods, they design a robust AGW benchmark that achieves state-of-the-art or comparable performance in single-modal or multi-modal person re-identification tasks. Additionally, a new evaluation metric (mlNP) is introduced to measure the cost of finding all the correct matches, which is crucial for practical applications in person re-identification. Overall, this survey provides a comprehensive understanding of the current state of deep learning for person re-identification and offers insights into future research directions.