Mitigating Hallucinations in Large Vision-Language Models with Internal Fact-based Contrastive Decoding

TL;DR

This paper introduces Internal Fact-based Contrastive Decoding (IFCD), a model-agnostic inference technique that reduces hallucinations in large vision-language models by leveraging their own internal representations, leading to improved accuracy.

Contribution

The paper proposes a novel inference-time method, IFCD, that mitigates hallucinations in LVLMs without additional training or external data, based on internal representation analysis.

Findings

Significantly reduces object and attribute hallucinations.

Achieves 9% accuracy improvement on POPE dataset.

Achieves 8% accuracy improvement on MME hallucination subset.

Abstract

Large Visual Language Models (LVLMs) integrate visual and linguistic modalities, exhibiting exceptional performance across various multimodal tasks. Nevertheless, LVLMs remain vulnerable to the issue of object hallucinations. Previous efforts to mitigate this issue focus on supervised fine-tuning (SFT) or incorporating external knowledge, both of which entail significant costs related to training and the acquisition of external data. To address these challenges, we propose a novel model-agnostic approach termed Internal Fact-based Contrastive Decoding (IFCD), designed to mitigate and suppress hallucinations during the inference process of LVLMs by exploiting the LVLMs' own hallucinations. IFCD is grounded in experimental observations that alterations to the LVLMs' internal representations tend to amplify hallucinations caused by language bias. By contrasting disturbed distribution, IFCD calibrates the LVLMs' output and effectively removes the hallucinatory logits from the final predictions. Experimental results validate that IFCD significantly alleviates both object-level and attribute-level hallucinations while achieving an average 9% accuracy improvement on POPE and 8% accuracy improvement on MME object hallucinations subset compared with direct decoding, respectively.