On the Credibility of Backdoor Attacks Against Object Detectors in the Physical World
Authors:
Bao Gia Doan,
Dang Quang Nguyen,
Callum Lindquist,
Paul Montague,
Tamas Abraham,
Olivier De Vel,
Seyit Camtepe,
Salil S. Kanhere,
Ehsan Abbasnejad,
Damith C. Ranasinghe
Abstract:
Object detectors are vulnerable to backdoor attacks. In contrast to classifiers, detectors possess unique characteristics, architecturally and in task execution; often operating in challenging conditions, for instance, detecting traffic signs in autonomous cars. But, our knowledge dominates attacks against classifiers and tests in the "digital domain".
To address this critical gap, we conducted…
▽ More
Object detectors are vulnerable to backdoor attacks. In contrast to classifiers, detectors possess unique characteristics, architecturally and in task execution; often operating in challenging conditions, for instance, detecting traffic signs in autonomous cars. But, our knowledge dominates attacks against classifiers and tests in the "digital domain".
To address this critical gap, we conducted an extensive empirical study targeting multiple detector architectures and two challenging detection tasks in real-world settings: traffic signs and vehicles. Using the diverse, methodically collected videos captured from driving cars and flying drones, incorporating physical object trigger deployments in authentic scenes, we investigated the viability of physical object-triggered backdoor attacks in application settings.
Our findings revealed 8 key insights. Importantly, the prevalent "digital" data poisoning method for injecting backdoors into models does not lead to effective attacks against detectors in the real world, although proven effective in classification tasks. We construct a new, cost-efficient attack method, dubbed MORPHING, incorporating the unique nature of detection tasks; ours is remarkably successful in injecting physical object-triggered backdoors, even capable of poisoning triggers with clean label annotations or invisible triggers without diminishing the success of physical object triggered backdoors. We discovered that the defenses curated are ill-equipped to safeguard detectors against such attacks. To underscore the severity of the threat and foster further research, we, for the first time, release an extensive video test set of real-world backdoor attacks. Our study not only establishes the credibility and seriousness of this threat but also serves as a clarion call to the research community to advance backdoor defenses in the context of object detection.
△ Less
Submitted 30 October, 2024; v1 submitted 22 August, 2024;
originally announced August 2024.
Timing and Frequency Synchronization for 1-bit Massive MU-MIMO-OFDM Downlink
Authors:
Sven Jacobsson,
Carl Lindquist,
Giuseppe Durisi,
Thomas Eriksson,
Christoph Studer
Abstract:
We consider timing and frequency synchronization for the massive multiuser (MU) multiple-input multiple-output (MIMO) downlink where 1-bit digital-to-analog converters (DACs) are used at the base station (BS). We focus on the practically relevant scenario in which orthogonal-frequency division multiplexing (OFDM) is used to communicate over frequency-selective channels. Our contributions are twofo…
▽ More
We consider timing and frequency synchronization for the massive multiuser (MU) multiple-input multiple-output (MIMO) downlink where 1-bit digital-to-analog converters (DACs) are used at the base station (BS). We focus on the practically relevant scenario in which orthogonal-frequency division multiplexing (OFDM) is used to communicate over frequency-selective channels. Our contributions are twofold. First, we use Bussgang's theorem to analyze the impact on performance caused by timing and frequency offsets in the presence of 1-bit DACs at the BS. Second, we demonstrate the efficacy of the widely used Schmidl-Cox synchronization algorithm. Our results demonstrate that the 1-bit massive MU-MIMO-OFDM downlink is resilient against timing and frequency offsets.
△ Less
Submitted 19 May, 2019;
originally announced May 2019.
Periodic Modulation of Extraordinary Optical Transmission through Subwavelength Hole Arrays using Surrounding Bragg Mirrors
Authors:
Nathan C. Lindquist,
Antoine Lesuffleur,
Sang-Hyun Oh
Abstract:
The enhanced light transmission through an array of subwavelength holes surrounded by Bragg mirrors is studied, showing that the mirrors act to confine the surface plasmons associated with the Extraordinary Optical Transmission effect, forming a surface resonant cavity. The overall effect is increased light transmission intensity by more than a factor of three beyond the already enhanced transmi…
▽ More
The enhanced light transmission through an array of subwavelength holes surrounded by Bragg mirrors is studied, showing that the mirrors act to confine the surface plasmons associated with the Extraordinary Optical Transmission effect, forming a surface resonant cavity. The overall effect is increased light transmission intensity by more than a factor of three beyond the already enhanced transmission, independent of whether the Bragg mirrors are on the input or the output side of the incident light. The geometry of the Bragg mirror structures controls the enhancement, and can even reduce the transmission in half. By varying these geometric parameters, we were able to periodically modulate the transmission of light for specific wavelengths, consistent with the propagation and interference of surface plasmon waves in a resonant cavity. FDTD simulations and a wave propagation model verify this effect.
△ Less
Submitted 9 August, 2007;
originally announced August 2007.
