Hidden Backdoor Attack against Semantic Segmentation Models

Deep neural networks (DNNs) are vulnerable to the \emph{backdoor attack}, which intends to embed hidden backdoors in DNNs by poisoning training data. The attacked model behaves normally on benign samples, whereas its prediction will be changed to a particular target label if hidden backdoors are act...

Ausführliche Beschreibung

Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Li, Yiming, Li, Yanjie, Lv, Yalei, Jiang, Yong, Xia, Shu-Tao
Format: Artikel
Sprache:eng
Schlagworte:
Online-Zugang:Volltext bestellen
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
container_end_page
container_issue
container_start_page
container_title
container_volume
creator Li, Yiming
Li, Yanjie
Lv, Yalei
Jiang, Yong
Xia, Shu-Tao
description Deep neural networks (DNNs) are vulnerable to the \emph{backdoor attack}, which intends to embed hidden backdoors in DNNs by poisoning training data. The attacked model behaves normally on benign samples, whereas its prediction will be changed to a particular target label if hidden backdoors are activated. So far, backdoor research has mostly been conducted towards classification tasks. In this paper, we reveal that this threat could also happen in semantic segmentation, which may further endanger many mission-critical applications ($e.g.$, autonomous driving). Except for extending the existing attack paradigm to maliciously manipulate the segmentation models from the image-level, we propose a novel attack paradigm, the \emph{fine-grained attack}, where we treat the target label ($i.e.$, annotation) from the object-level instead of the image-level to achieve more sophisticated manipulation. In the annotation of poisoned samples generated by the fine-grained attack, only pixels of specific objects will be labeled with the attacker-specified target class while others are still with their ground-truth ones. Experiments show that the proposed methods can successfully attack semantic segmentation models by poisoning only a small proportion of training data. Our method not only provides a new perspective for designing novel attacks but also serves as a strong baseline for improving the robustness of semantic segmentation methods.
doi_str_mv 10.48550/arxiv.2103.04038
format Article
fullrecord <record><control><sourceid>arxiv_GOX</sourceid><recordid>TN_cdi_arxiv_primary_2103_04038</recordid><sourceformat>XML</sourceformat><sourcesystem>PC</sourcesystem><sourcerecordid>2103_04038</sourcerecordid><originalsourceid>FETCH-LOGICAL-a678-7ad5cc3861efc80ec1fbc4f4d0f252e6140ad7887ffa112645da5d7707d458b13</originalsourceid><addsrcrecordid>eNotj7FuwjAURb0wINoPYKp_IOE5tmNLnQC1pRKIAfboxc-OrBIHJVbV_n0p7XTPdHUOY0sBpbJawwrHr_hZVgJkCQqknbPnXSTyiW_QfdAwjHyd8w05dhjTlPnJ95hydDfoep8y5jgkfhjIX6YHNgt4mfzj_y7Y-fXlvN0V--Pb-3a9L7A2tjBI2jlpa-GDs-CdCK1TQRGESle-FgqQjLUmBBSiqpUm1GQMGFLatkIu2NPf7d2-uY6xx_G7-a1o7hXyB4G3QgY</addsrcrecordid><sourcetype>Open Access Repository</sourcetype><iscdi>true</iscdi><recordtype>article</recordtype></control><display><type>article</type><title>Hidden Backdoor Attack against Semantic Segmentation Models</title><source>arXiv.org</source><creator>Li, Yiming ; Li, Yanjie ; Lv, Yalei ; Jiang, Yong ; Xia, Shu-Tao</creator><creatorcontrib>Li, Yiming ; Li, Yanjie ; Lv, Yalei ; Jiang, Yong ; Xia, Shu-Tao</creatorcontrib><description>Deep neural networks (DNNs) are vulnerable to the \emph{backdoor attack}, which intends to embed hidden backdoors in DNNs by poisoning training data. The attacked model behaves normally on benign samples, whereas its prediction will be changed to a particular target label if hidden backdoors are activated. So far, backdoor research has mostly been conducted towards classification tasks. In this paper, we reveal that this threat could also happen in semantic segmentation, which may further endanger many mission-critical applications ($e.g.$, autonomous driving). Except for extending the existing attack paradigm to maliciously manipulate the segmentation models from the image-level, we propose a novel attack paradigm, the \emph{fine-grained attack}, where we treat the target label ($i.e.$, annotation) from the object-level instead of the image-level to achieve more sophisticated manipulation. In the annotation of poisoned samples generated by the fine-grained attack, only pixels of specific objects will be labeled with the attacker-specified target class while others are still with their ground-truth ones. Experiments show that the proposed methods can successfully attack semantic segmentation models by poisoning only a small proportion of training data. Our method not only provides a new perspective for designing novel attacks but also serves as a strong baseline for improving the robustness of semantic segmentation methods.</description><identifier>DOI: 10.48550/arxiv.2103.04038</identifier><language>eng</language><subject>Computer Science - Artificial Intelligence ; Computer Science - Computer Vision and Pattern Recognition ; Computer Science - Cryptography and Security</subject><creationdate>2021-03</creationdate><rights>http://arxiv.org/licenses/nonexclusive-distrib/1.0</rights><oa>free_for_read</oa><woscitedreferencessubscribed>false</woscitedreferencessubscribed></display><links><openurl>$$Topenurl_article</openurl><openurlfulltext>$$Topenurlfull_article</openurlfulltext><thumbnail>$$Tsyndetics_thumb_exl</thumbnail><link.rule.ids>228,230,776,881</link.rule.ids><linktorsrc>$$Uhttps://arxiv.org/abs/2103.04038$$EView_record_in_Cornell_University$$FView_record_in_$$GCornell_University$$Hfree_for_read</linktorsrc><backlink>$$Uhttps://doi.org/10.48550/arXiv.2103.04038$$DView paper in arXiv$$Hfree_for_read</backlink></links><search><creatorcontrib>Li, Yiming</creatorcontrib><creatorcontrib>Li, Yanjie</creatorcontrib><creatorcontrib>Lv, Yalei</creatorcontrib><creatorcontrib>Jiang, Yong</creatorcontrib><creatorcontrib>Xia, Shu-Tao</creatorcontrib><title>Hidden Backdoor Attack against Semantic Segmentation Models</title><description>Deep neural networks (DNNs) are vulnerable to the \emph{backdoor attack}, which intends to embed hidden backdoors in DNNs by poisoning training data. The attacked model behaves normally on benign samples, whereas its prediction will be changed to a particular target label if hidden backdoors are activated. So far, backdoor research has mostly been conducted towards classification tasks. In this paper, we reveal that this threat could also happen in semantic segmentation, which may further endanger many mission-critical applications ($e.g.$, autonomous driving). Except for extending the existing attack paradigm to maliciously manipulate the segmentation models from the image-level, we propose a novel attack paradigm, the \emph{fine-grained attack}, where we treat the target label ($i.e.$, annotation) from the object-level instead of the image-level to achieve more sophisticated manipulation. In the annotation of poisoned samples generated by the fine-grained attack, only pixels of specific objects will be labeled with the attacker-specified target class while others are still with their ground-truth ones. Experiments show that the proposed methods can successfully attack semantic segmentation models by poisoning only a small proportion of training data. Our method not only provides a new perspective for designing novel attacks but also serves as a strong baseline for improving the robustness of semantic segmentation methods.</description><subject>Computer Science - Artificial Intelligence</subject><subject>Computer Science - Computer Vision and Pattern Recognition</subject><subject>Computer Science - Cryptography and Security</subject><fulltext>true</fulltext><rsrctype>article</rsrctype><creationdate>2021</creationdate><recordtype>article</recordtype><sourceid>GOX</sourceid><recordid>eNotj7FuwjAURb0wINoPYKp_IOE5tmNLnQC1pRKIAfboxc-OrBIHJVbV_n0p7XTPdHUOY0sBpbJawwrHr_hZVgJkCQqknbPnXSTyiW_QfdAwjHyd8w05dhjTlPnJ95hydDfoep8y5jgkfhjIX6YHNgt4mfzj_y7Y-fXlvN0V--Pb-3a9L7A2tjBI2jlpa-GDs-CdCK1TQRGESle-FgqQjLUmBBSiqpUm1GQMGFLatkIu2NPf7d2-uY6xx_G7-a1o7hXyB4G3QgY</recordid><startdate>20210306</startdate><enddate>20210306</enddate><creator>Li, Yiming</creator><creator>Li, Yanjie</creator><creator>Lv, Yalei</creator><creator>Jiang, Yong</creator><creator>Xia, Shu-Tao</creator><scope>AKY</scope><scope>GOX</scope></search><sort><creationdate>20210306</creationdate><title>Hidden Backdoor Attack against Semantic Segmentation Models</title><author>Li, Yiming ; Li, Yanjie ; Lv, Yalei ; Jiang, Yong ; Xia, Shu-Tao</author></sort><facets><frbrtype>5</frbrtype><frbrgroupid>cdi_FETCH-LOGICAL-a678-7ad5cc3861efc80ec1fbc4f4d0f252e6140ad7887ffa112645da5d7707d458b13</frbrgroupid><rsrctype>articles</rsrctype><prefilter>articles</prefilter><language>eng</language><creationdate>2021</creationdate><topic>Computer Science - Artificial Intelligence</topic><topic>Computer Science - Computer Vision and Pattern Recognition</topic><topic>Computer Science - Cryptography and Security</topic><toplevel>online_resources</toplevel><creatorcontrib>Li, Yiming</creatorcontrib><creatorcontrib>Li, Yanjie</creatorcontrib><creatorcontrib>Lv, Yalei</creatorcontrib><creatorcontrib>Jiang, Yong</creatorcontrib><creatorcontrib>Xia, Shu-Tao</creatorcontrib><collection>arXiv Computer Science</collection><collection>arXiv.org</collection></facets><delivery><delcategory>Remote Search Resource</delcategory><fulltext>fulltext_linktorsrc</fulltext></delivery><addata><au>Li, Yiming</au><au>Li, Yanjie</au><au>Lv, Yalei</au><au>Jiang, Yong</au><au>Xia, Shu-Tao</au><format>journal</format><genre>article</genre><ristype>JOUR</ristype><atitle>Hidden Backdoor Attack against Semantic Segmentation Models</atitle><date>2021-03-06</date><risdate>2021</risdate><abstract>Deep neural networks (DNNs) are vulnerable to the \emph{backdoor attack}, which intends to embed hidden backdoors in DNNs by poisoning training data. The attacked model behaves normally on benign samples, whereas its prediction will be changed to a particular target label if hidden backdoors are activated. So far, backdoor research has mostly been conducted towards classification tasks. In this paper, we reveal that this threat could also happen in semantic segmentation, which may further endanger many mission-critical applications ($e.g.$, autonomous driving). Except for extending the existing attack paradigm to maliciously manipulate the segmentation models from the image-level, we propose a novel attack paradigm, the \emph{fine-grained attack}, where we treat the target label ($i.e.$, annotation) from the object-level instead of the image-level to achieve more sophisticated manipulation. In the annotation of poisoned samples generated by the fine-grained attack, only pixels of specific objects will be labeled with the attacker-specified target class while others are still with their ground-truth ones. Experiments show that the proposed methods can successfully attack semantic segmentation models by poisoning only a small proportion of training data. Our method not only provides a new perspective for designing novel attacks but also serves as a strong baseline for improving the robustness of semantic segmentation methods.</abstract><doi>10.48550/arxiv.2103.04038</doi><oa>free_for_read</oa></addata></record>
fulltext fulltext_linktorsrc
identifier DOI: 10.48550/arxiv.2103.04038
ispartof
issn
language eng
recordid cdi_arxiv_primary_2103_04038
source arXiv.org
subjects Computer Science - Artificial Intelligence
Computer Science - Computer Vision and Pattern Recognition
Computer Science - Cryptography and Security
title Hidden Backdoor Attack against Semantic Segmentation Models
url https://sfx.bib-bvb.de/sfx_tum?ctx_ver=Z39.88-2004&ctx_enc=info:ofi/enc:UTF-8&ctx_tim=2025-02-08T19%3A07%3A01IST&url_ver=Z39.88-2004&url_ctx_fmt=infofi/fmt:kev:mtx:ctx&rfr_id=info:sid/primo.exlibrisgroup.com:primo3-Article-arxiv_GOX&rft_val_fmt=info:ofi/fmt:kev:mtx:journal&rft.genre=article&rft.atitle=Hidden%20Backdoor%20Attack%20against%20Semantic%20Segmentation%20Models&rft.au=Li,%20Yiming&rft.date=2021-03-06&rft_id=info:doi/10.48550/arxiv.2103.04038&rft_dat=%3Carxiv_GOX%3E2103_04038%3C/arxiv_GOX%3E%3Curl%3E%3C/url%3E&disable_directlink=true&sfx.directlink=off&sfx.report_link=0&rft_id=info:oai/&rft_id=info:pmid/&rfr_iscdi=true