AdvShadow: Camouflaged Adversarial Attacks via Conditional Diffusion Model-Generated Shadows

Abstract Adversarial examples pose a significant challenge to the robustness of deep neural networks(DNNs).Traditional adversarial attacks often lack semantic features and concealment. This paper proposesAdvShadow, an adversarial example generation algorithm based on a conditional diffusion model. Ad-vShadow dynamically adjusts shadow perturbations to attack sensitive regions of images, generatingvisually realistic shadow effects that deceive deep models. Our approach ensures that the adversarialshadows seamlessly integrate with image features, enhancing attack concealment and effectiveness.Experiments on multiple datasets demonstrate that AdvShadow achieves high attack success rateswhile maintaining high image quality, as evidenced by improved structural similarity index (SSIM)and peak signal-to-noise ratio (PSNR) scores. This study not only provides a novel adversarial attackmethod but also sheds light on the robustness of deep learning models against stealthy attacks. Theraw data Oxford-IIIt-Pet required to reproduce the above findings are available to download fromOxford-IIIT Pet Dataset. Code will be released in AdvShadow