Person Re-identification in Identity Regression Space

• DOI: 10.1007/s11263-018-1105-3
• Publication Date: 2018-07-26T23:55:15Z
• Authors: Hanxiao Wang; Xiatian Zhu; Shaogang Gong; Tao Xiang
• Keywords: Benchmarking; Identification

Most existing person re-identification (re-id) methods are unsuitable for real-world deployment due to two reasons: Unscalability large population size, and Inadaptability over time. In this work, we present a unified solution address both problems. Specifically, propose construct an identity regression space (IRS) based on embedding different training identities (classes) formulate re-id as problem solved by in the IRS. The IRS approach is characterised closed-form with high learning efficiency inherent incremental capability human-in-the-loop. Extensive experiments four benchmarking datasets (VIPeR, CUHK01, CUHK03 Market-1501) show that model not only outperforms state-of-the-art methods, but also more scalable size rapidly updating actively selecting informative samples reduced human labelling effort.