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

RGB-D object recognition systems improve their predictive performances byfusing color and depth information, outperforming neural network architecturesthat rely solely on colors. While RGB-D systems are expected to be more robustto adversarial examples than RGB-only systems, they have also been proven to behighly vulnerable. Their robustness is similar even when the adversarialexamples are generated by altering only the original images' colors. Differentworks highlighted the vulnerability of RGB-D systems; however, there is alacking of technical explanations for this weakness. Hence, in our work, webridge this gap by investigating the learned deep representation of RGB-Dsystems, discovering that color features make the function learned by thenetwork more complex and, thus, more sensitive to small perturbations. Tomitigate this problem, we propose a defense based on a detection mechanism thatmakes RGB-D systems more robust against adversarial examples. We empiricallyshow that this defense improves the performances of RGB-D systems againstadversarial examples even when they are computed ad-hoc to circumvent thisdetection mechanism, and that is also more effective than adversarial training.

Hardening RGB-D Object Recognition Systems against Adversarial Patch Attacks