English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Deep learning models have been shown to be vulnerable to recent backdoorattacks. A backdoored model behaves normally for inputs containing noattacker-secretly-chosen trigger and maliciously for inputs with the trigger.To date, backdoor attacks and countermeasures mainly focus on imageclassification tasks. And most of them are implemented in the digital worldwith digital triggers. Besides the classification tasks, object detectionsystems are also considered as one of the basic foundations of computer visiontasks. However, there is no investigation and understanding of the backdoorvulnerability of the object detector, even in the digital world with digitaltriggers. For the first time, this work demonstrates that existing objectdetectors are inherently susceptible to physical backdoor attacks. We use anatural T-shirt bought from a market as a trigger to enable the cloakingeffect--the person bounding-box disappears in front of the object detector. Weshow that such a backdoor can be implanted from two exploitable attackscenarios into the object detector, which is outsourced or fine-tuned through apretrained model. We have extensively evaluated three popular object detectionalgorithms: anchor-based Yolo-V3, Yolo-V4, and anchor-free CenterNet. Buildingupon 19 videos shot in real-world scenes, we confirm that the backdoor attackis robust against various factors: movement, distance, angle, non-rigiddeformation, and lighting. Specifically, the attack success rate (ASR) in mostvideos is 100% or close to it, while the clean data accuracy of the backdooredmodel is the same as its clean counterpart. The latter implies that it isinfeasible to detect the backdoor behavior merely through a validation set. Theaveraged ASR still remains sufficiently high to be 78% in the transfer learningattack scenarios evaluated on CenterNet. See the demo video onhttps://youtu.be/Q3HOF4OobbY.

Dangerous Cloaking: Natural Trigger based Backdoor Attacks on Object Detectors in the Physical World