Abstract

In more recent years the development of computer vision has advanced to be more comprehensive than in the past, with newer applications ranging from autonomous vehicles to security systems. The main application I will be talking about throughout this paper is an object detection algorithm called YOLO (You only look once), this algorithm is particularly significant due to their real-time performance of being able to identify and localize objects within an image in quick timing. However, the strength of these computer vision models is increasingly challenged by adversarial attacks, which manipulate the computer's vision to block a certain part of the image out or to confuse the software into thinking it is something else. Throughout this research paper I will be explaining how YOLO works and how shaped adversarial patches are able to fool this algorithm.

Faculty Advisor/Mentor

Lida Haghnegahdar

Document Type

Paper

Disciplines

Cybersecurity

DOI

10.25776/vs1t-y532

Publication Date

4-17-2025

Upload File

wf_yes

Included in

Cybersecurity Commons

Share

COinS
 

Shaped Adversarial Patches and the Ability They Hold

In more recent years the development of computer vision has advanced to be more comprehensive than in the past, with newer applications ranging from autonomous vehicles to security systems. The main application I will be talking about throughout this paper is an object detection algorithm called YOLO (You only look once), this algorithm is particularly significant due to their real-time performance of being able to identify and localize objects within an image in quick timing. However, the strength of these computer vision models is increasingly challenged by adversarial attacks, which manipulate the computer's vision to block a certain part of the image out or to confuse the software into thinking it is something else. Throughout this research paper I will be explaining how YOLO works and how shaped adversarial patches are able to fool this algorithm.