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

Object detection has been widely used in many safety-critical tasks, such asautonomous driving. However, its vulnerability to adversarial examples has notbeen sufficiently studied, especially under the practical scenario of black-boxattacks, where the attacker can only access the query feedback of predictedbounding-boxes and top-1 scores returned by the attacked model. Compared withblack-box attack to image classification, there are two main challenges inblack-box attack to detection. Firstly, even if one bounding-box issuccessfully attacked, another sub-optimal bounding-box may be detected nearthe attacked bounding-box. Secondly, there are multiple bounding-boxes, leadingto very high attack cost. To address these challenges, we propose a ParallelRectangle Flip Attack (PRFA) via random search. We explain the differencebetween our method with other attacks in Fig.~\ref{fig1}. Specifically, wegenerate perturbations in each rectangle patch to avoid sub-optimal detectionnear the attacked region. Besides, utilizing the observation that adversarialperturbations mainly locate around objects' contours and critical points underwhite-box attacks, the search space of attacked rectangles is reduced toimprove the attack efficiency. Moreover, we develop a parallel mechanism ofattacking multiple rectangles simultaneously to further accelerate the attackprocess. Extensive experiments demonstrate that our method can effectively andefficiently attack various popular object detectors, including anchor-based andanchor-free, and generate transferable adversarial examples.

Parallel Rectangle Flip Attack: A Query-based Black-box Attack against Object Detection