Stefan Milenkovic
Hardening RGB-D object recognition systems against adversarial patch attacks
Demetrio, Luca;Demontis, Ambra
;Biggio, Battista;Roli, FabioLast
2023-01-01
Abstract
RGB-D object recognition systems improve their predictive performances by fusing color and depth information, outperforming neural network architectures that rely solely on colors. While RGB-D systems are expected to be more robust to adversarial examples than RGB-only systems, they have also been proven to be highly vulnerable. Their robustness is similar even when the adversarial examples are generated by altering only the original images' colors. Different works highlighted the vulnerability of RGB-D systems; however, there is a lacking of technical explanations for this weakness. Hence, in our work, we bridge this gap by investigating the learned deep representation of RGB-D systems, discovering that color features make the function learned by the network more complex and, thus, more sensitive to small perturbations. To mitigate this problem, we propose a defense based on a detection mechanism that makes RGB-D systems more robust against adversarial examples. We empirically show that this defense improves the performances of RGB-D systems against adversarial examples even when they are computed ad-hoc to circumvent this detection mechanism, and that is also more effective than adversarial training.File | Size | Format | |
---|---|---|---|
1-s2.0-S0020025523012860-main.pdf Solo gestori archivio
Type: versione editoriale
Size 2.79 MB
Format Adobe PDF
|
2.79 MB | Adobe PDF | & nbsp; View / Open Request a copy |
Hardening RGB-D_preprint.pdf open access
Type: versione pre-print
Size 1.72 MB
Format Adobe PDF
|
1.72 MB | Adobe PDF | View/Open |
Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.