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

arXiv:2403.18715v1 Announce Type: new
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 …

abstract adept application arxiv contents cs.ai cs.cl cs.cv cs.mm decision decoding generated hallucinations however inputs language language models making multimodal rate responses text type vision vision-language models visual