ZINA: Multimodal Fine-grained Hallucination Detection and Editing

Fuente: arXiv
Salvato in:
Dettagli Bibliografici
Autori principali: Wada, Yuiga, Matsuda, Kazuki, Sugiura, Komei, Neubig, Graham
Natura: Preprint
Pubblicazione: 2025
Soggetti:
Accesso online:
Tags: Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
_version_ 1866918428860743680
author Wada, Yuiga
Matsuda, Kazuki
Sugiura, Komei
Neubig, Graham
author_facet Wada, Yuiga
Matsuda, Kazuki
Sugiura, Komei
Neubig, Graham
contents Multimodal Large Language Models (MLLMs) often generate hallucinations, where the output deviates from the visual content. Given that these hallucinations can take diverse forms, detecting hallucinations at a fine-grained level is essential for comprehensive evaluation and analysis. To this end, we propose a novel task of multimodal fine-grained hallucination detection and editing for MLLMs. Moreover, we propose ZINA, a novel method that identifies hallucinated spans at a fine-grained level, classifies their error types into six categories, and suggests appropriate refinements. To train and evaluate models for this task, we construct VisionHall, a dataset comprising 6.9k outputs from twelve MLLMs manually annotated by 211 annotators, and 20k synthetic samples generated using a graph-based method that captures dependencies among error types. We demonstrated that ZINA outperformed existing methods, including GPT-4o and Llama-3.2, in both detection and editing tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2506_13130
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle ZINA: Multimodal Fine-grained Hallucination Detection and Editing
Wada, Yuiga
Matsuda, Kazuki
Sugiura, Komei
Neubig, Graham
Computer Vision and Pattern Recognition
Artificial Intelligence
Computation and Language
Multimodal Large Language Models (MLLMs) often generate hallucinations, where the output deviates from the visual content. Given that these hallucinations can take diverse forms, detecting hallucinations at a fine-grained level is essential for comprehensive evaluation and analysis. To this end, we propose a novel task of multimodal fine-grained hallucination detection and editing for MLLMs. Moreover, we propose ZINA, a novel method that identifies hallucinated spans at a fine-grained level, classifies their error types into six categories, and suggests appropriate refinements. To train and evaluate models for this task, we construct VisionHall, a dataset comprising 6.9k outputs from twelve MLLMs manually annotated by 211 annotators, and 20k synthetic samples generated using a graph-based method that captures dependencies among error types. We demonstrated that ZINA outperformed existing methods, including GPT-4o and Llama-3.2, in both detection and editing tasks.
title ZINA: Multimodal Fine-grained Hallucination Detection and Editing
topic Computer Vision and Pattern Recognition
Artificial Intelligence
Computation and Language
url https://arxiv.org/abs/2506.13130