CutPaste&Find: Efficient Multimodal Hallucination Detector with Visual-aid Knowledge Base

Cong-Duy Nguyen, Xiaobao Wu, Duc Anh Vu, Shuai Zhao, Thong Nguyen, Anh Tuan Luu * · 2025

Read CutPaste&Find: Efficient Multimodal Hallucination Detector with Visual-aid Knowledge Base by Cong-Duy Nguyen, Xiaobao Wu, Duc Anh Vu, Shuai Zhao, Thong Nguyen, Anh Tuan Luu * (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2502-12591.txt.

Description

Cong-Duy Nguyen and co-authors present a cheap way to catch vision-language models inventing things that are not in a picture, in a 2025 paper. Such models sometimes describe objects that never appear, a failure called object hallucination. Existing detectors lean on costly calls to big models, repeated many times per image. Their framework, CutPaste&Find, instead uses ready-made vision and language tools in a checking pipeline, with no extra training.

At its heart sits a knowledge base pairing things with their typical traits and example images. Candidate claims from a model's description are cut out and matched against that base. A scaling tweak fixes the awkward fact that even true pairs can score low on raw likeness. Tests on standard benchmarks showed strong detection without the heavy cost, the authors report, making checks feasible at scale.

Catalog record

CREATORCong-Duy Nguyen, Xiaobao Wu, Duc Anh Vu, Shuai Zhao, Thong Nguyen, Anh Tuan Luu *
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE35 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2502-12591

Browse the full library · About the Tower of Babel