VL-Uncertainty: Detecting Hallucination in Large Vision-Language Model via Uncertainty Estimation

TL;DR

VL-Uncertainty is a novel framework that detects hallucinations in large vision-language models by measuring response uncertainty through semantic clustering, outperforming existing methods across multiple benchmarks.

Contribution

Introduces the first uncertainty-based hallucination detection method for LVLMs that does not rely on ground-truth annotations.

Findings

Outperforms baseline methods in hallucination detection

Effective across diverse LVLMs and benchmarks

Utilizes semantic clustering and entropy for uncertainty measurement

Abstract

Given the higher information load processed by large vision-language models (LVLMs) compared to single-modal LLMs, detecting LVLM hallucinations requires more human and time expense, and thus rise a wider safety concerns. In this paper, we introduce VL-Uncertainty, the first uncertainty-based framework for detecting hallucinations in LVLMs. Different from most existing methods that require ground-truth or pseudo annotations, VL-Uncertainty utilizes uncertainty as an intrinsic metric. We measure uncertainty by analyzing the prediction variance across semantically equivalent but perturbed prompts, including visual and textual data. When LVLMs are highly confident, they provide consistent responses to semantically equivalent queries. However, when uncertain, the responses of the target LVLM become more random. Considering semantically similar answers with different wordings, we cluster LVLM responses based on their semantic content and then calculate the cluster distribution entropy as the uncertainty measure to detect hallucination. Our extensive experiments on 10 LVLMs across four benchmarks, covering both free-form and multi-choice tasks, show that VL-Uncertainty significantly outperforms strong baseline methods in hallucination detection.