A CMA‐ES‐Based Adversarial Attack Against Black‐Box Object Detectors

Object detection is one of the essential tasks computer vision. detectors based on deep neural network have been used more and widely in safe-sensitive applications, like face recognition, video surveillance, autonomous driving, other tasks. It has proved that object are vulnerable to adversarial attacks. We propose a novel black-box attack method, which can successfully regression-based region-based detectors. introduce methods reduce search dimensions, dimension optimization problems number queries by using Covariance matrix adaptation Evolution strategy (CMA-ES) as primary method generate examples. Our only adds perturbations box achieve precise attack. proposed hide specified with an success rate 86% average 5, 124, all objects 74% 6, 154. work illustrates effectiveness CMA-ES examples proves vulnerability against