Design and Interpretation of Universal Adversarial Patches in Face Detection

TL;DR

This paper explores universal adversarial patches for face detection, revealing their face-like appearance and proposing optimization methods to effectively deceive state-of-the-art detectors across various scenarios.

Contribution

It introduces a novel interpretation of face-like adversarial patches and develops optimization algorithms for their automatic design targeting different attack goals.

Findings

Patches appear face-like and can reliably suppress face detection.

Proposed methods effectively deceive detectors on real-world datasets.

Patches transfer across different detection frameworks.

Abstract

We consider universal adversarial patches for faces -- small visual elements whose addition to a face image reliably destroys the performance of face detectors. Unlike previous work that mostly focused on the algorithmic design of adversarial examples in terms of improving the success rate as an attacker, in this work we show an interpretation of such patches that can prevent the state-of-the-art face detectors from detecting the real faces. We investigate a phenomenon: patches designed to suppress real face detection appear face-like. This phenomenon holds generally across different initialization, locations, scales of patches, backbones, and state-of-the-art face detection frameworks. We propose new optimization-based approaches to automatic design of universal adversarial patches for varying goals of the attack, including scenarios in which true positives are suppressed without introducing false positives. Our proposed algorithms perform well on real-world datasets, deceiving state-of-the-art face detectors in terms of multiple precision/recall metrics and transferability.