Contextual Attribution Maps-Guided Transferable Adversarial Attack for 3D Object Detection

The study of LiDAR-based 3D object detection and its robustness under adversarial attacks has achieved great progress. However, existing adversarial attack methods mainly focus on the targeted object, which destroys the integrity of the object and makes the attack easy to perceive. In this work, we...

Full description

Saved in:
Bibliographic Details
Main Authors: Mumuxin Cai, Xupeng Wang, Ferdous Sohel, Hang Lei
Format: Article
Language:English
Published: MDPI AG 2024-11-01
Series:Remote Sensing
Subjects:
Online Access:https://www.mdpi.com/2072-4292/16/23/4409
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1850060344319279104
author Mumuxin Cai
Xupeng Wang
Ferdous Sohel
Hang Lei
author_facet Mumuxin Cai
Xupeng Wang
Ferdous Sohel
Hang Lei
author_sort Mumuxin Cai
collection DOAJ
description The study of LiDAR-based 3D object detection and its robustness under adversarial attacks has achieved great progress. However, existing adversarial attack methods mainly focus on the targeted object, which destroys the integrity of the object and makes the attack easy to perceive. In this work, we propose a novel adversarial attack against deep 3D object detection models named the contextual attribution maps-guided attack (CAMGA). Based on the combinations of subregions in the context area and their impact on the prediction results, contextual attribution maps can be generated. An attribution map exposes the influence of individual subregions in the context area on the detection results and narrows down the scope of the adversarial attack. Subsequently, perturbations are generated under the guidance of a dual loss, which is proposed to suppress the detection results and maintain visual imperception simultaneously. The experimental results proved that the CAMGA method achieved an attack success rate of over 68% on three large-scale datasets and 83% on the KITTI dataset. Meanwhile, the CAMGA has a transfer attack success rate of at least 50% against all four victim detectors, as they all overly rely on contextual information.
format Article
id doaj-art-9e7743f1dd40498892511afad06fd8b1
institution DOAJ
issn 2072-4292
language English
publishDate 2024-11-01
publisher MDPI AG
record_format Article
series Remote Sensing
spelling doaj-art-9e7743f1dd40498892511afad06fd8b12025-08-20T02:50:36ZengMDPI AGRemote Sensing2072-42922024-11-011623440910.3390/rs16234409Contextual Attribution Maps-Guided Transferable Adversarial Attack for 3D Object DetectionMumuxin Cai0Xupeng Wang1Ferdous Sohel2Hang Lei3School of Information and Software Engineering, The University of Electronic Science and Technology of China, Chengdu 610054, ChinaLaboratory of Intelligent Collaborative Computing, The University of Electronic Science and Technology of China, Chengdu 611731, ChinaSchool of Information Technology, Murdoch University, Murdoch, WA 6150, AustraliaSchool of Information and Software Engineering, The University of Electronic Science and Technology of China, Chengdu 610054, ChinaThe study of LiDAR-based 3D object detection and its robustness under adversarial attacks has achieved great progress. However, existing adversarial attack methods mainly focus on the targeted object, which destroys the integrity of the object and makes the attack easy to perceive. In this work, we propose a novel adversarial attack against deep 3D object detection models named the contextual attribution maps-guided attack (CAMGA). Based on the combinations of subregions in the context area and their impact on the prediction results, contextual attribution maps can be generated. An attribution map exposes the influence of individual subregions in the context area on the detection results and narrows down the scope of the adversarial attack. Subsequently, perturbations are generated under the guidance of a dual loss, which is proposed to suppress the detection results and maintain visual imperception simultaneously. The experimental results proved that the CAMGA method achieved an attack success rate of over 68% on three large-scale datasets and 83% on the KITTI dataset. Meanwhile, the CAMGA has a transfer attack success rate of at least 50% against all four victim detectors, as they all overly rely on contextual information.https://www.mdpi.com/2072-4292/16/23/4409adversarial attack3D object detectioncontextual informationtransferable attackattribution map
spellingShingle Mumuxin Cai
Xupeng Wang
Ferdous Sohel
Hang Lei
Contextual Attribution Maps-Guided Transferable Adversarial Attack for 3D Object Detection
Remote Sensing
adversarial attack
3D object detection
contextual information
transferable attack
attribution map
title Contextual Attribution Maps-Guided Transferable Adversarial Attack for 3D Object Detection
title_full Contextual Attribution Maps-Guided Transferable Adversarial Attack for 3D Object Detection
title_fullStr Contextual Attribution Maps-Guided Transferable Adversarial Attack for 3D Object Detection
title_full_unstemmed Contextual Attribution Maps-Guided Transferable Adversarial Attack for 3D Object Detection
title_short Contextual Attribution Maps-Guided Transferable Adversarial Attack for 3D Object Detection
title_sort contextual attribution maps guided transferable adversarial attack for 3d object detection
topic adversarial attack
3D object detection
contextual information
transferable attack
attribution map
url https://www.mdpi.com/2072-4292/16/23/4409
work_keys_str_mv AT mumuxincai contextualattributionmapsguidedtransferableadversarialattackfor3dobjectdetection
AT xupengwang contextualattributionmapsguidedtransferableadversarialattackfor3dobjectdetection
AT ferdoussohel contextualattributionmapsguidedtransferableadversarialattackfor3dobjectdetection
AT hanglei contextualattributionmapsguidedtransferableadversarialattackfor3dobjectdetection