Meta Generative Attack on Person Reidentification

TL;DR

This paper introduces a meta generative attack method for person re-identification that enhances transferability across different models and datasets by using mask generation and meta learning techniques.

Contribution

It proposes a novel meta generative attack approach that improves attack transferability in cross-dataset cross-model scenarios for person re-identification.

Findings

Achieves better attack transferability across models and datasets.

Demonstrates superior performance on Market-1501, DukeMTMC-reID, and MSMT-17 datasets.

Outperforms existing attack methods in experiments.

Abstract

Adversarial attacks have been recently investigated in person re-identification. These attacks perform well under cross dataset or cross model setting. However, the challenges present in cross-dataset cross-model scenario does not allow these models to achieve similar accuracy. To this end, we propose our method with the goal of achieving better transferability against different models and across datasets. We generate a mask to obtain better performance across models and use meta learning to boost the generalizability in the challenging cross-dataset cross-model setting. Experiments on Market-1501, DukeMTMC-reID and MSMT-17 demonstrate favorable results compared to other attacks.