Vision Language Models Map Logos to Text via Semantic Entanglement in the Visual Projector

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Li, Sifan, Chen, Hongkai, Cai, Yujun, Ye, Qingwen, Chen, Liyang, Yuan, Junsong, Wang, Yiwei
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866915553580417024
author Li, Sifan
Chen, Hongkai
Cai, Yujun
Ye, Qingwen
Chen, Liyang
Yuan, Junsong
Wang, Yiwei
author_facet Li, Sifan
Chen, Hongkai
Cai, Yujun
Ye, Qingwen
Chen, Liyang
Yuan, Junsong
Wang, Yiwei
contents Vision Language Models (VLMs) have achieved impressive progress in multimodal reasoning; yet, they remain vulnerable to hallucinations, where outputs are not grounded in visual evidence. In this paper, we investigate a previously overlooked setting: logo hallucination, where models generate brand names or textual content despite logos containing no visible words. Using curated splits of pure symbols, hybrids, and text-bearing logos, as well as the challenging Hard-60 subset, we systematically measure hallucination across leading VLMs. We further probe robustness through nine structured perturbations and show that hallucinations persist even under strong distortions, with occlusion exposing the sharpest weaknesses. Embedding-level analysis with open-weight LLaVA demonstrates that hallucination is tied to a small subset of projector dimensions, and targeted ablation substantially reduces errors while preserving OCR accuracy. Together, these findings reveal that VLMs often rely on symbolic priors rather than genuine glyph perception, particularly for iconic circular logos, and that projector subspaces play a decisive role in this failure mode. Our work contributes both a novel diagnostic lens and actionable mitigation insights, highlighting projector disentanglement and OCR-guided decoding as promising directions for building more trustworthy multimodal systems.
format Preprint
id arxiv_https___arxiv_org_abs_2510_12287
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Vision Language Models Map Logos to Text via Semantic Entanglement in the Visual Projector
Li, Sifan
Chen, Hongkai
Cai, Yujun
Ye, Qingwen
Chen, Liyang
Yuan, Junsong
Wang, Yiwei
Computer Vision and Pattern Recognition
Computation and Language
Vision Language Models (VLMs) have achieved impressive progress in multimodal reasoning; yet, they remain vulnerable to hallucinations, where outputs are not grounded in visual evidence. In this paper, we investigate a previously overlooked setting: logo hallucination, where models generate brand names or textual content despite logos containing no visible words. Using curated splits of pure symbols, hybrids, and text-bearing logos, as well as the challenging Hard-60 subset, we systematically measure hallucination across leading VLMs. We further probe robustness through nine structured perturbations and show that hallucinations persist even under strong distortions, with occlusion exposing the sharpest weaknesses. Embedding-level analysis with open-weight LLaVA demonstrates that hallucination is tied to a small subset of projector dimensions, and targeted ablation substantially reduces errors while preserving OCR accuracy. Together, these findings reveal that VLMs often rely on symbolic priors rather than genuine glyph perception, particularly for iconic circular logos, and that projector subspaces play a decisive role in this failure mode. Our work contributes both a novel diagnostic lens and actionable mitigation insights, highlighting projector disentanglement and OCR-guided decoding as promising directions for building more trustworthy multimodal systems.
title Vision Language Models Map Logos to Text via Semantic Entanglement in the Visual Projector
topic Computer Vision and Pattern Recognition
Computation and Language
url https://arxiv.org/abs/2510.12287