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

In recent times, monocular depth estimation (MDE) has experienced significantadvancements in performance, largely attributed to the integration ofinnovative architectures, i.e., convolutional neural networks (CNNs) andTransformers. Nevertheless, the susceptibility of these models to adversarialattacks has emerged as a noteworthy concern, especially in domains where safetyand security are paramount. This concern holds particular weight for MDE due toits critical role in applications like autonomous driving and roboticnavigation, where accurate scene understanding is pivotal. To assess thevulnerability of CNN-based depth prediction methods, recent work tries todesign adversarial patches against MDE. However, the existing approaches fallshort of inducing a comprehensive and substantially disruptive impact on thevision system. Instead, their influence is partial and confined to specificlocal areas. These methods lead to erroneous depth predictions only within theoverlapping region with the input image, without considering thecharacteristics of the target object, such as its size, shape, and position. Inthis paper, we introduce a novel adversarial patch named APARATE. This patchpossesses the ability to selectively undermine MDE in two distinct ways: bydistorting the estimated distances or by creating the illusion of an objectdisappearing from the perspective of the autonomous system. Notably, APARATE isdesigned to be sensitive to the shape and scale of the target object, and itsinfluence extends beyond immediate proximity. APARATE, results in a mean depthestimation error surpassing $0.5$, significantly impacting as much as $99\%$ ofthe targeted region when applied to CNN-based MDE models. Furthermore, ityields a significant error of $0.34$ and exerts substantial influence over$94\%$ of the target region in the context of Transformer-based MDE.

APARATE: Adaptive Adversarial Patch for CNN-based Monocular Depth Estimation for Autonomous Navigation