Saved in:
Bibliographic Details
Main Authors: Cai, Hongyi, Rahman, Mohammad Mahdinur, Dong, Mingkang, Pu, Muxin, Alqaily, Moqyad, Li, Jie, Li, Xinfeng, Shen, Jialie, Qiu, Meikang, Wen, Qingsong
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2508.00445
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911473377214464
author Cai, Hongyi
Rahman, Mohammad Mahdinur
Dong, Mingkang
Pu, Muxin
Alqaily, Moqyad
Li, Jie
Li, Xinfeng
Shen, Jialie
Qiu, Meikang
Wen, Qingsong
author_facet Cai, Hongyi
Rahman, Mohammad Mahdinur
Dong, Mingkang
Pu, Muxin
Alqaily, Moqyad
Li, Jie
Li, Xinfeng
Shen, Jialie
Qiu, Meikang
Wen, Qingsong
contents Text-to-Image (T2I) models generate high-quality images but are vulnerable to malicious backdoor attacks that inject harmful biases (e.g., trigger-activated gender or racial stereotypes). Existing debiasing methods, often designed for natural statistical biases, struggle with these deliberately and subtly injected attacks. We propose AutoDebias, a framework that automatically identifies and mitigates these malicious biases in T2I models without prior knowledge of the specific attack types. Specifically, AutoDebias leverages vision-language models to detect trigger-activated visual patterns and constructs neutralization guides by generating counter-prompts. These guides drive a CLIP-guided training process that breaks the harmful associations while preserving the original model's image quality and diversity. Unlike methods designed for natural bias, AutoDebias effectively addresses subtle, injected stereotypes and multiple interacting attacks. We evaluate the framework on a new benchmark covering 17 distinct backdoor scenarios, including challenging cases where multiple backdoors co-exist. AutoDebias detects malicious patterns with 91.6% accuracy and reduces the backdoor success rate from 90% to negligible levels, while preserving the visual fidelity of the original model.
format Preprint
id arxiv_https___arxiv_org_abs_2508_00445
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle AutoDebias: Automated Framework for Debiasing Text-to-Image Models
Cai, Hongyi
Rahman, Mohammad Mahdinur
Dong, Mingkang
Pu, Muxin
Alqaily, Moqyad
Li, Jie
Li, Xinfeng
Shen, Jialie
Qiu, Meikang
Wen, Qingsong
Computer Vision and Pattern Recognition
Text-to-Image (T2I) models generate high-quality images but are vulnerable to malicious backdoor attacks that inject harmful biases (e.g., trigger-activated gender or racial stereotypes). Existing debiasing methods, often designed for natural statistical biases, struggle with these deliberately and subtly injected attacks. We propose AutoDebias, a framework that automatically identifies and mitigates these malicious biases in T2I models without prior knowledge of the specific attack types. Specifically, AutoDebias leverages vision-language models to detect trigger-activated visual patterns and constructs neutralization guides by generating counter-prompts. These guides drive a CLIP-guided training process that breaks the harmful associations while preserving the original model's image quality and diversity. Unlike methods designed for natural bias, AutoDebias effectively addresses subtle, injected stereotypes and multiple interacting attacks. We evaluate the framework on a new benchmark covering 17 distinct backdoor scenarios, including challenging cases where multiple backdoors co-exist. AutoDebias detects malicious patterns with 91.6% accuracy and reduces the backdoor success rate from 90% to negligible levels, while preserving the visual fidelity of the original model.
title AutoDebias: Automated Framework for Debiasing Text-to-Image Models
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2508.00445