EnsemHalDet: Robust VLM Hallucination Detection via Ensemble of Internal State Detectors

TL;DR

EnsemHalDet introduces an ensemble approach leveraging multiple internal VLM representations to improve hallucination detection accuracy and robustness in multimodal tasks.

Contribution

It proposes a novel ensemble framework that combines various internal signals of VLMs for more effective hallucination detection.

Findings

EnsemHalDet outperforms prior methods in AUC across multiple datasets.

Ensembling diverse internal signals enhances detection robustness.

The approach is effective across different VQA datasets and models.

Abstract

Vision-Language Models (VLMs) excel at multimodal tasks, but they remain vulnerable to hallucinations that are factually incorrect or ungrounded in the input image. Recent work suggests that hallucination detection using internal representations is more efficient and accurate than approaches that rely solely on model outputs. However, existing internal-representation-based methods typically rely on a single representation or detector, limiting their ability to capture diverse hallucination signals. In this paper, we propose EnsemHalDet, an ensemble-based hallucination detection framework that leverages multiple internal representations of VLMs, including attention outputs and hidden states. EnsemHalDet trains independent detectors for each representation and combines them through ensemble learning. Experimental results across multiple VQA datasets and VLMs show that EnsemHalDet consistently outperforms prior methods and single-detector models in terms of AUC. These results demonstrate that ensembling diverse internal signals significantly improves robustness in multimodal hallucination detection.