Vision Language Models Map Logos to Text via Semantic Entanglement in the Visual Projector
This work addresses a vulnerability in VLMs for multimodal reasoning, offering diagnostic and mitigation insights to improve trustworthiness, though it is incremental as it builds on existing VLM analysis.
The paper tackled the problem of logo hallucination in Vision Language Models (VLMs), where models generate brand names from logos without visible text, and found that hallucinations persist under distortions and are linked to specific projector dimensions, with targeted ablation reducing errors while maintaining OCR accuracy.
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.