Who Brings the Frisbee: Probing Hidden Hallucination Factors in Large Vision-Language Model via Causality Analysis

TL;DR

This paper investigates the hidden causes of hallucinations in large vision-language models using causality analysis, proposing a probing system to identify and mitigate factors leading to non-existent visual elements.

Contribution

It introduces a novel causal probing approach to identify hidden factors causing hallucinations in LVLMs and demonstrates effective interventions to reduce such hallucinations.

Findings

Causality analysis reveals key hidden factors inducing hallucinations.

Interventions based on the analysis significantly reduce hallucinated outputs.

Potential to edit network internals to further minimize hallucinations.

Abstract

Recent advancements in large vision-language models (LVLM) have significantly enhanced their ability to comprehend visual inputs alongside natural language. However, a major challenge in their real-world application is hallucination, where LVLMs generate non-existent visual elements, eroding user trust. The underlying mechanism driving this multimodal hallucination is poorly understood. Minimal research has illuminated whether contexts such as sky, tree, or grass field involve the LVLM in hallucinating a frisbee. We hypothesize that hidden factors, such as objects, contexts, and semantic foreground-background structures, induce hallucination. This study proposes a novel causal approach: a hallucination probing system to identify these hidden factors. By analyzing the causality between images, text prompts, and network saliency, we systematically explore interventions to block these factors. Our experimental findings show that a straightforward technique based on our analysis can significantly reduce hallucinations. Additionally, our analyses indicate the potential to edit network internals to minimize hallucinated outputs.