Poison as Cure: Visual Noise for Mitigating Object Hallucinations in LVMs

TL;DR

This paper introduces a novel visual adversarial perturbation method that applies optimized visual noise to large vision-language models, significantly reducing object hallucinations and improving factual accuracy without modifying the models.

Contribution

The paper presents a new adversarial noise technique to mitigate hallucinations in LVMs, enhancing their reliability without altering the underlying models.

Findings

Consistently reduces object hallucinations across 8 state-of-the-art LVMs

Enhances factual grounding and reduces knowledge bias

Validated through extensive experiments

Abstract

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.