Mitigating Hallucinations in Large Vision-Language Models with Instruction Contrastive Decoding

TL;DR

This paper introduces Instruction Contrastive Decoding (ICD), a novel method that significantly reduces hallucinations in large vision-language models, improving their accuracy and perception capabilities during multimodal tasks.

Contribution

The paper proposes ICD, a new decoding technique that contrasts distributions to mitigate hallucinations in LVLMs, enhancing their reliability and perception accuracy.

Findings

ICD reduces object and attribute hallucinations in LVLMs.

ICD improves LVLM performance on discriminative and generative benchmarks.

ICD enhances the perception and recognition capabilities of LVLMs.

Abstract

Large Vision-Language Models (LVLMs) are increasingly adept at generating contextually detailed and coherent responses from visual inputs. However, their application in multimodal decision-making and open-ended generation is hindered by a notable rate of hallucinations, where generated text inaccurately represents the visual contents. To address this issue, this paper introduces the Instruction Contrastive Decoding (ICD) method, a novel approach designed to reduce hallucinations during LVLM inference. Our method is inspired by our observation that what we call disturbance instructions significantly exacerbate hallucinations in multimodal fusion modules. ICD contrasts distributions from standard and instruction disturbance, thereby increasing alignment uncertainty and effectively subtracting hallucinated concepts from the original distribution. Through comprehensive experiments on discriminative benchmarks (POPE and MME) and a generative benchmark (LLaVa-Bench), we demonstrate that ICD significantly mitigates both object-level and attribute-level hallucinations. Moreover, our method not only addresses hallucinations but also significantly enhances the general perception and recognition capabilities of LVLMs.