Metric learning with trace-norm regularization for person re-identification

Yao, Bohuai; Zhao, Zhicheng; Li, Kai

doi:10.1109/icip.2014.7025494

Cited by 2 publications

(3 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…We compared the proposed KRML methods using linear,

χ^{2}

, and Hellinger kernels (denoted as KRML‐L, KRML‐

χ^{2}

, and KRML‐H) with several state‐of‐the‐art methods [1–3, 5–8, 11, 12] on two challenging and widely used person re‐identification datasets, i.e. CAVIAR [13] and 3DPeS [14].…”

Section: Resultsmentioning

confidence: 99%

“…CAVIAR [13] and 3DPeS [14]. We follow the experimental protocols in [1][2][3][4][5], and report the average cumulative match characteristic (CMC) curves and the normalised area under CMC (nAUC). In our experiments, we adopt a three-way data split for the parameter tuning, i.e.…”

Section: Problem Formulationmentioning

confidence: 99%

“…Introduction: Distance metric, which measures closeness between samples, is critical in many machine learning and computer vision applications. In recent years, a bunch of works resort to metric learning to improve the performance of person re-identification that aims to identify the target among a set of gallery candidates across non-overlapping camera views [1][2][3][4][5][6]. However, person re-identification remains a very challenging task due to the large intra-class dissimilarity and small interclass dissimilarity caused by unconstrained variations in viewpoint, illumination, background clutter, occlusion, pose and so on.…”

mentioning

confidence: 99%

See 2 more Smart Citations

Kernelised reference‐wise metric learning

Luo

et al. 2017

Electron. lett.

View full text Add to dashboard Cite

Unlike the doublet or triplet constraints, a novel kernelised referencewise metric learning is proposed by constructing reference-wise constraints, which contain similarity information of each sample to all reference samples. After selecting several representative training samples as the reference set, the training data are first mapped into a projected space by the reference sample matrix. Then the problem of metric learning is casted as a multi-label classification problem under the reference-wise constraints, in which an l 2-norm regularised least squares canonical correlation analysis can be used. Besides, the formulation is generalised to kernelised version for further boosting the performance. Experiments on two benchmark person re-identification datasets demonstrate that the approach clearly outperforms the stateof-the-art methods.

show abstract

“…We compared the proposed KRML methods using linear,

χ^{2}

, and Hellinger kernels (denoted as KRML‐L, KRML‐

χ^{2}

, and KRML‐H) with several state‐of‐the‐art methods [1–3, 5–8, 11, 12] on two challenging and widely used person re‐identification datasets, i.e. CAVIAR [13] and 3DPeS [14].…”

Section: Resultsmentioning

confidence: 99%