UHP Detection: LVLMs have their Unique Hallucination Pattern in the Consistency Space

2026-08-04

AI Summary

This work addresses the challenge of hallucinations in large vision-language models (LVLMs), which often generate responses inconsistent with visual content. Existing black-box detection methods, relying on a single consistency metric, struggle to capture the diverse manifestations of such hallucinations. To overcome this limitation, the authors propose the Uncertainty-based Hallucination Profiling (UHP) framework, which models hallucinations as structured uncertainty patterns defined by image/text perturbations and logical polarity. UHP constructs four complementary consistency groups and leverages intra- and inter-group features to train a lightweight classifier for structured hallucination identification. Experiments demonstrate that UHP achieves state-of-the-art performance, improving AUC-ROC by up to 18.72% and AUC-PR by 20.07% on the AMBER and PhD benchmarks, outperforming both black-box and white-box baselines. Moreover, the identified hallucination patterns exhibit strong cross-dataset generalization.

Abstract

Large vision--language models (LVLMs) demonstrate strong multimodal reasoning capabilities but remain prone to hallucination, where model predictions are not grounded in visual evidence. Existing black-box hallucination detection methods estimate uncertainty through a single consistency metric, implicitly assuming that model uncertainty can be adequately characterized by a single measure. However, hallucinations exhibit diverse manifestations of uncertainty across different behavioral probes, making a single measure insufficient to characterize their underlying behavior. We propose \emph{Unique Hallucination Pattern (UHP) Detection}, a fully black-box framework that models hallucination as a structured uncertainty pattern defined by two axes: perturbation modality (image vs.\ text) and logical polarity (a statement vs.\ its negation). Their intersection produces four complementary consistency groups that capture distinct manifestations of model uncertainty, from which both within-group and between-group features are extracted to train a lightweight classifier. Through comprehensive experiments on AMBER and PhD across three LVLMs, UHP Detection consistently outperforms prior black-box and white-box baselines, with improvements of up to $+18.72\%$ AUC-ROC and $+20.07\%$ AUC-PR over the strongest black-box methods. Extensive ablation studies demonstrate that each consistency group contributes complementary information and that their combination forms a structured hallucination pattern. Furthermore, cross-dataset evaluation shows that this learned pattern generalizes across benchmarks, indicating that hallucination behavior reflects a model-specific consistency pattern. \textbf{Code is publicly available at} https://github.com/amirezzati/uhpdet.

Problem

• hallucination detection
• large vision-language models
• consistency metrics
• model uncertainty
• black-box methods

Innovation

• hallucination detection
• large vision-language models
• consistency space
• structured uncertainty
• black-box evaluation