KDAT: Inherent Adversarial Robustness via Knowledge Distillation with Adversarial Tuning for Object Detection Models
Proceedings of the AAAI Conference on Artificial Intelligence · 11.04.2025 · 39 (5) · 4598-4606
DOI: 10.1609/aaai.v39i5.32485 ↗Kurz gefasst
Fachartikel in der Zeitschrift Proceedings of the AAAI Conference on Artificial Intelligence, online erschienen am 11.04.2025. Verfasst von Yarin Yerushalmi Levi und 7 weiteren Autor:innen. Die Arbeit wurde bisher 6-mal zitiert (Crossref).
Abstract (Originalsprache)
Adversarial patches pose a significant threat to computer vision models' integrity, decreasing the accuracy of various tasks, including object detection (OD). Most existing OD defenses exhibit a trade-off between enhancing the model's adversarial robustness and maintaining its performance on benign images. We propose KDAT (knowledge distillation with adversarial tuning), a novel mechanism that enhances the robustness of an OD model without compromising its performance on benign images or its inference time. Our method combines the knowledge distillation (KD) technique with the adversarial tuning concept to teach the model to match the predictions of adversarial images with those of their corresponding benign ones. To match these predictions, we designed four unique loss components, allowing the student model to effectively distill the knowledge of different features from various parts of the teacher model. Our extensive evaluation on the COCO and INRIA datasets demonstrates KDAT's ability to improve the performance of Faster R-CNN and DETR on benign images by 2-4 mAP% and adversarial examples by 10-15 mAP%, outperforming other state-of-the-art (SOTA) defenses. Furthermore, our additional physical evaluation on the Superstore dataset demonstrates KDAT's SOTA adversarial robustness against printed patches (improvement of 22 mAP% compared to the undefended model).
Zitieren
APA
Levi, Y. Y., Grolman, E., Yankelev, I., Giloni, A., Hofman, O., Shimizu, T., Shabtai, A., & Elovici, Y. (2025). KDAT: Inherent Adversarial Robustness via Knowledge Distillation with Adversarial Tuning for Object Detection Models. Proceedings of the AAAI Conference on Artificial Intelligence, 39(5), 4598-4606. https://doi.org/10.1609/aaai.v39i5.32485
Vancouver
Levi YY, Grolman E, Yankelev I, Giloni A, Hofman O, Shimizu T, et al. KDAT: Inherent Adversarial Robustness via Knowledge Distillation with Adversarial Tuning for Object Detection Models. Proceedings of the AAAI Conference on Artificial Intelligence. 2025;39(5):4598-4606. doi:10.1609/aaai.v39i5.32485
BibTeX
@article{Levi2025,
author = {Levi, Yarin Yerushalmi and Grolman, Edita and Yankelev, Idan and Giloni, Amit and Hofman, Omer and Shimizu, Toshiya and Shabtai, Asaf and Elovici, Yuval},
title = {KDAT: Inherent Adversarial Robustness via Knowledge Distillation with Adversarial Tuning for Object Detection Models},
journal = {Proceedings of the AAAI Conference on Artificial Intelligence},
year = {2025},
volume = {39},
number = {5},
pages = {4598--4606},
doi = {10.1609/aaai.v39i5.32485}
}
EUVIMED