Adversarial Attacks of Vision Tasks in the Past 10 Years: A Survey

TL;DR

This survey comprehensively reviews adversarial attacks on vision tasks over the past decade, highlighting traditional and LVLM-specific threats, their characteristics, and implications for improving system robustness.

Contribution

It provides an integrated analysis of traditional and LVLM adversarial attacks, addressing gaps in taxonomy, evaluation frameworks, and attack motivation insights.

Findings

Unified insights into attack transferability and generalization.

Detailed evaluation frameworks for adversarial robustness.

Connections and distinctions between traditional and LVLM attacks.

Abstract

With the advent of Large Vision-Language Models (LVLMs), new attack vectors, such as cognitive bias, prompt injection, and jailbreaking, have emerged. Understanding these attacks promotes system robustness improvement and neural networks demystification. However, existing surveys often target attack taxonomy and lack in-depth analysis like 1) unified insights into adversariality, transferability, and generalization; 2) detailed evaluations framework; 3) motivation-driven attack categorizations; and 4) an integrated perspective on both traditional and LVLM attacks. This article addresses these gaps by offering a thorough summary of traditional and LVLM adversarial attacks, emphasizing their connections and distinctions, and providing actionable insights for future research.