Saved in:
Bibliographic Details
Main Authors: Zhang, Kejia, Tao, Keda, Tang, Jiasheng, Wang, Huan
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2501.19164
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866909503325208576
author Zhang, Kejia
Tao, Keda
Tang, Jiasheng
Wang, Huan
author_facet Zhang, Kejia
Tao, Keda
Tang, Jiasheng
Wang, Huan
contents Large vision-language models (LVMs) extend large language models (LLMs) with visual perception capabilities, enabling them to process and interpret visual information. A major challenge compromising their reliability is object hallucination that LVMs may generate plausible but factually inaccurate information. We propose a novel visual adversarial perturbation (VAP) method to mitigate this hallucination issue. VAP alleviates LVM hallucination by applying strategically optimized visual noise without altering the base model. Our approach formulates hallucination suppression as an optimization problem, leveraging adversarial strategies to generate beneficial visual perturbations that enhance the model's factual grounding and reduce parametric knowledge bias. Extensive experimental results demonstrate that our method consistently reduces object hallucinations across 8 state-of-the-art LVMs, validating its efficacy across diverse evaluations.
format Preprint
id arxiv_https___arxiv_org_abs_2501_19164
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Poison as Cure: Visual Noise for Mitigating Object Hallucinations in LVMs
Zhang, Kejia
Tao, Keda
Tang, Jiasheng
Wang, Huan
Computer Vision and Pattern Recognition
Large vision-language models (LVMs) extend large language models (LLMs) with visual perception capabilities, enabling them to process and interpret visual information. A major challenge compromising their reliability is object hallucination that LVMs may generate plausible but factually inaccurate information. We propose a novel visual adversarial perturbation (VAP) method to mitigate this hallucination issue. VAP alleviates LVM hallucination by applying strategically optimized visual noise without altering the base model. Our approach formulates hallucination suppression as an optimization problem, leveraging adversarial strategies to generate beneficial visual perturbations that enhance the model's factual grounding and reduce parametric knowledge bias. Extensive experimental results demonstrate that our method consistently reduces object hallucinations across 8 state-of-the-art LVMs, validating its efficacy across diverse evaluations.
title Poison as Cure: Visual Noise for Mitigating Object Hallucinations in LVMs
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2501.19164