HBReID: Harder Batch for Re-identification

TL;DR

This paper introduces a hard batch mining method for person re-identification that selects globally hardest samples and incorporates scene invariance, achieving state-of-the-art results on MSMT17.

Contribution

It proposes a novel hard batch mining technique and an adversarial scene removal module to improve re-identification performance.

Findings

Surpasses previous methods on MSMT17 dataset

Effectively mines globally hardest samples within batches

Learns scene-invariant feature representations

Abstract

Triplet loss is a widely adopted loss function in ReID task which pulls the hardest positive pairs close and pushes the hardest negative pairs far away. However, the selected samples are not the hardest globally, but the hardest only in a mini-batch, which will affect the performance. In this report, a hard batch mining method is proposed to mine the hardest samples globally to make triplet harder. More specifically, the most similar classes are selected into a same mini-batch so that the similar classes could be pushed further away. Besides, an adversarial scene removal module composed of a scene classifier and an adversarial loss is used to learn scene invariant feature representations. Experiments are conducted on dataset MSMT17 to prove the effectiveness, and our method surpasses all of the previous methods and sets state-of-the-art result.