The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
Grâce à l’excellente capacité d’apprentissage des réseaux neuronaux à convolution profonde (CNN), les méthodes basées sur les CNN ont obtenu un grand succès dans les tâches de vision par ordinateur et de reconnaissance d’images. Cependant, il s’est avéré que ces méthodes comportent souvent des vulnérabilités inhérentes, ce qui nous rend prudents quant aux risques potentiels liés à leur utilisation pour des applications réelles telles que la conduite autonome. Pour révéler de telles vulnérabilités, nous proposons une méthode d'attaque simultanée de l'estimation de la profondeur monoculaire et de l'estimation du flux optique, qui sont toutes deux des tâches courantes basées sur l'intelligence artificielle qui font l'objet d'études approfondies pour des scénarios de conduite autonome. Notre méthode peut générer un correctif contradictoire qui peut tromper simultanément les méthodes d'estimation de profondeur monoculaire basées sur CNN et d'estimation de flux optique en plaçant simplement le correctif dans les images d'entrée. À notre connaissance, il s'agit du premier travail permettant de réaliser des attaques de correctifs simultanées sur deux ou plusieurs CNN développés pour des tâches différentes.
Koichiro YAMANAKA
Nagoya University
Keita TAKAHASHI
Nagoya University
Toshiaki FUJII
Nagoya University
Ryuraroh MATSUMOTO
Tokyo Institute of Technology
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copier
Koichiro YAMANAKA, Keita TAKAHASHI, Toshiaki FUJII, Ryuraroh MATSUMOTO, "Simultaneous Attack on CNN-Based Monocular Depth Estimation and Optical Flow Estimation" in IEICE TRANSACTIONS on Information,
vol. E104-D, no. 5, pp. 785-788, May 2021, doi: 10.1587/transinf.2021EDL8004.
Abstract: Thanks to the excellent learning capability of deep convolutional neural networks (CNNs), CNN-based methods have achieved great success in computer vision and image recognition tasks. However, it has turned out that these methods often have inherent vulnerabilities, which makes us cautious of the potential risks of using them for real-world applications such as autonomous driving. To reveal such vulnerabilities, we propose a method of simultaneously attacking monocular depth estimation and optical flow estimation, both of which are common artificial-intelligence-based tasks that are intensively investigated for autonomous driving scenarios. Our method can generate an adversarial patch that can fool CNN-based monocular depth estimation and optical flow estimation methods simultaneously by simply placing the patch in the input images. To the best of our knowledge, this is the first work to achieve simultaneous patch attacks on two or more CNNs developed for different tasks.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.2021EDL8004/_p
Copier
@ARTICLE{e104-d_5_785,
author={Koichiro YAMANAKA, Keita TAKAHASHI, Toshiaki FUJII, Ryuraroh MATSUMOTO, },
journal={IEICE TRANSACTIONS on Information},
title={Simultaneous Attack on CNN-Based Monocular Depth Estimation and Optical Flow Estimation},
year={2021},
volume={E104-D},
number={5},
pages={785-788},
abstract={Thanks to the excellent learning capability of deep convolutional neural networks (CNNs), CNN-based methods have achieved great success in computer vision and image recognition tasks. However, it has turned out that these methods often have inherent vulnerabilities, which makes us cautious of the potential risks of using them for real-world applications such as autonomous driving. To reveal such vulnerabilities, we propose a method of simultaneously attacking monocular depth estimation and optical flow estimation, both of which are common artificial-intelligence-based tasks that are intensively investigated for autonomous driving scenarios. Our method can generate an adversarial patch that can fool CNN-based monocular depth estimation and optical flow estimation methods simultaneously by simply placing the patch in the input images. To the best of our knowledge, this is the first work to achieve simultaneous patch attacks on two or more CNNs developed for different tasks.},
keywords={},
doi={10.1587/transinf.2021EDL8004},
ISSN={1745-1361},
month={May},}
Copier
TY - JOUR
TI - Simultaneous Attack on CNN-Based Monocular Depth Estimation and Optical Flow Estimation
T2 - IEICE TRANSACTIONS on Information
SP - 785
EP - 788
AU - Koichiro YAMANAKA
AU - Keita TAKAHASHI
AU - Toshiaki FUJII
AU - Ryuraroh MATSUMOTO
PY - 2021
DO - 10.1587/transinf.2021EDL8004
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E104-D
IS - 5
JA - IEICE TRANSACTIONS on Information
Y1 - May 2021
AB - Thanks to the excellent learning capability of deep convolutional neural networks (CNNs), CNN-based methods have achieved great success in computer vision and image recognition tasks. However, it has turned out that these methods often have inherent vulnerabilities, which makes us cautious of the potential risks of using them for real-world applications such as autonomous driving. To reveal such vulnerabilities, we propose a method of simultaneously attacking monocular depth estimation and optical flow estimation, both of which are common artificial-intelligence-based tasks that are intensively investigated for autonomous driving scenarios. Our method can generate an adversarial patch that can fool CNN-based monocular depth estimation and optical flow estimation methods simultaneously by simply placing the patch in the input images. To the best of our knowledge, this is the first work to achieve simultaneous patch attacks on two or more CNNs developed for different tasks.
ER -