![[2602.22144] NoLan: Mitigating Object Hallucinations in Large Vision-Language Models via Dynamic Suppression of Language Priors](https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png){kind=logo}
[2602.22144] NoLan: Mitigating Object Hallucinations in Large Vision-Language Models via Dynamic Suppression of Language Priors
Summary
The paper presents NoLan, a framework aimed at reducing object hallucinations in Large Vision-Language Models (LVLMs) by dynamically suppressing language priors, thus improving model accuracy.
Why It Matters
Object hallucination in LVLMs can lead to inaccurate outputs, impacting applications in AI and machine learning. Understanding and mitigating this issue is crucial for enhancing the reliability of AI systems that integrate vision and language, making this research significant for developers and researchers in the field.
Key Takeaways
- Object hallucinations in LVLMs are mainly caused by strong language priors from the decoder.
- NoLan offers a training-free solution to mitigate hallucinations effectively.
- Experimental results show significant accuracy improvements in models like LLaVA-1.5 and Qwen-VL.
- The framework can be applied across various tasks and LVLMs.
- Public availability of the code encourages further research and application.
Computer Science > Computer Vision and Pattern Recognition arXiv:2602.22144 (cs) [Submitted on 25 Feb 2026] Title:NoLan: Mitigating Object Hallucinations in Large Vision-Language Models via Dynamic Suppression of Language Priors Authors:Lingfeng Ren, Weihao Yu, Runpeng Yu, Xinchao Wang View a PDF of the paper titled NoLan: Mitigating Object Hallucinations in Large Vision-Language Models via Dynamic Suppression of Language Priors, by Lingfeng Ren and 3 other authors View PDF HTML (experimental) Abstract:Object hallucination is a critical issue in Large Vision-Language Models (LVLMs), where outputs include objects that do not appear in the input image. A natural question arises from this phenomenon: Which component of the LVLM pipeline primarily contributes to object hallucinations? The vision encoder to perceive visual information, or the language decoder to generate text responses? In this work, we strive to answer this question through designing a systematic experiment to analyze the roles of the vision encoder and the language decoder in hallucination generation. Our observations reveal that object hallucinations are predominantly associated with the strong priors from the language decoder. Based on this finding, we propose a simple and training-free framework, No-Language-Hallucination Decoding, NoLan, which refines the output distribution by dynamically suppressing language priors, modulated based on the output distribution difference between multimodal and text-only i...
