Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Nakada, Shota, Saito, Kazuhiro, Ishikawa, Yuchi, Munakata, Hokuto, Komatsu, Tatsuya, Kondo, Masayoshi
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:https://arxiv.org/abs/2510.25225
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866909875764723712
author Nakada, Shota
Saito, Kazuhiro
Ishikawa, Yuchi
Munakata, Hokuto
Komatsu, Tatsuya
Kondo, Masayoshi
author_facet Nakada, Shota
Saito, Kazuhiro
Ishikawa, Yuchi
Munakata, Hokuto
Komatsu, Tatsuya
Kondo, Masayoshi
contents We propose a novel task, hallucination localization in video captioning, which aims to identify hallucinations in video captions at the span level (i.e. individual words or phrases). This allows for a more detailed analysis of hallucinations compared to existing sentence-level hallucination detection task. To establish a benchmark for hallucination localization, we construct HLVC-Dataset, a carefully curated dataset created by manually annotating 1,167 video-caption pairs from VideoLLM-generated captions. We further implement a VideoLLM-based baseline method and conduct quantitative and qualitative evaluations to benchmark current performance on hallucination localization.
format Preprint
id arxiv_https___arxiv_org_abs_2510_25225
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Hallucination Localization in Video Captioning
Nakada, Shota
Saito, Kazuhiro
Ishikawa, Yuchi
Munakata, Hokuto
Komatsu, Tatsuya
Kondo, Masayoshi
Multimedia
We propose a novel task, hallucination localization in video captioning, which aims to identify hallucinations in video captions at the span level (i.e. individual words or phrases). This allows for a more detailed analysis of hallucinations compared to existing sentence-level hallucination detection task. To establish a benchmark for hallucination localization, we construct HLVC-Dataset, a carefully curated dataset created by manually annotating 1,167 video-caption pairs from VideoLLM-generated captions. We further implement a VideoLLM-based baseline method and conduct quantitative and qualitative evaluations to benchmark current performance on hallucination localization.
title Hallucination Localization in Video Captioning
topic Multimedia
url https://arxiv.org/abs/2510.25225