Fooling Object Detectors: Adversarial Attacks by Half-Neighbor Masks

TL;DR

This paper introduces a novel adversarial attack method called HNM-PGD that effectively fools object detection systems under strict constraints, demonstrating high competitiveness in a major competition.

Contribution

The paper presents a new adversarial attack technique specifically designed for object detectors, expanding the scope of adversarial research beyond classifiers.

Findings

Achieved top 1% ranking in CIKM 2020 AnalytiCup with the proposed attack.

Developed a strong perturbation method applicable to various object detectors.

Released code for reproducibility and further research.

Abstract

Although there are a great number of adversarial attacks on deep learning based classifiers, how to attack object detection systems has been rarely studied. In this paper, we propose a Half-Neighbor Masked Projected Gradient Descent (HNM-PGD) based attack, which can generate strong perturbation to fool different kinds of detectors under strict constraints. We also applied the proposed HNM-PGD attack in the CIKM 2020 AnalytiCup Competition, which was ranked within the top 1% on the leaderboard. We release the code at https://github.com/YanghaoZYH/HNM-PGD.