Distribution Context Aware Loss for Person Re-identification

TL;DR

This paper introduces a Distribution Context Aware (DCA) loss for person re-identification that considers the distribution context in feature space, improving clustering and matching accuracy over traditional pair-wise loss functions.

Contribution

The paper proposes a novel DCA loss that incorporates distribution context into deep metric learning for person re-identification, addressing limitations of existing pair-wise loss methods.

Findings

Outperforms baseline methods on Market-1501, DukeMTMC-reID, and MSMT17 datasets.

Enhances clustering quality in feature space.

Achieves state-of-the-art or competitive results in person re-identification.

Abstract

To learn the optimal similarity function between probe and gallery images in Person re-identification, effective deep metric learning methods have been extensively explored to obtain discriminative feature embedding. However, existing metric loss like triplet loss and its variants always emphasize pair-wise relations but ignore the distribution context in feature space, leading to inconsistency and sub-optimal. In fact, the similarity of one pair not only decides the match of this pair, but also has potential impacts on other sample pairs. In this paper, we propose a novel Distribution Context Aware (DCA) loss based on triplet loss to combine both numerical similarity and relation similarity in feature space for better clustering. Extensive experiments on three benchmarks including Market-1501, DukeMTMC-reID and MSMT17, evidence the favorable performance of our method against the corresponding baseline and other state-of-the-art methods.