Re-identification by neighborhood structure metric learning
作者：
Highlights：
• NSML is proposed to tackle the data variability and sparsity problem in person re-identification.
• NSML learns discriminative dissimilarities on the novel neighborhood structure manifold.
• NSML solves the non-convex optimization problem by the new cutting-surface approach.
• The effectiveness, robustness, efficiency, stability, and generalizability of NSML are experimentally validated.
摘要
Highlights•NSML is proposed to tackle the data variability and sparsity problem in person re-identification.•NSML learns discriminative dissimilarities on the novel neighborhood structure manifold.•NSML solves the non-convex optimization problem by the new cutting-surface approach.•The effectiveness, robustness, efficiency, stability, and generalizability of NSML are experimentally validated.
论文关键词：Re-identification,Metric learning,Neighborhood structure manifold
论文评审过程：Received 8 July 2015, Revised 20 May 2016, Accepted 1 August 2016, Available online 5 August 2016, Version of Record 16 August 2016.
论文官网地址：https://doi.org/10.1016/j.patcog.2016.08.001
