@inproceedings{huang-etal-2026-mitigating-object,
title = "Mitigating Object Hallucination in {MLLM}s with Global-to-Local Visual Attention Correction",
author = "Huang, Xingxing and
Yang, Zhiguo and
Pi, Jiacheng and
Xu, Dongsheng and
Zhong, Ruizhi and
Xu, Peipei and
Ruan, Wenjie",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.emnlp-main.2272/",
pages = "51671--51689",
ISBN = "979-8-89176-459-0",
abstract = "Multimodal Large Language Models (MLLMs) have achieved impressive progress in vision-language understanding, yet they remain susceptible to object hallucination, which refers to generating descriptions of objects that do not exist in the input image. Through comprehensive analysis, we uncover a fundamental trade-off in MLLMs: strengthening visual attention enhances grounding completeness but also increases hallucinations, especially additive hallucinations that appear in the trailing part of the generated sentence. To alleviate this trade-off, we propose **GLVAC**, **G**lobal-to-**L**ocal **V**isual **A**ttention **C**orrection, a training-free dual-path inference framework that adaptively combines global conservatism and local enhancement in visual attention. The global path maintains stable and grounded predictions under moderate visual attention, while the local correction branch is dynamically activated for high-uncertainty tokens identified via entropy-based estimation. A confidence-guided gating mechanism then determines whether to adopt the correction output, ensuring safe and reliable refinement. GLVAC can be seamlessly integrated into existing MLLMs without retraining or external modules, effectively reducing object hallucinations while maintaining perceptual completeness across diverse benchmarks and model families. Our code is available on https://github.com/Trusted-LLM/GLVAC."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="huang-etal-2026-mitigating-object">
<titleInfo>
<title>Mitigating Object Hallucination in MLLMs with Global-to-Local Visual Attention Correction</title>
</titleInfo>
<name type="personal">
<namePart type="given">Xingxing</namePart>
<namePart type="family">Huang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zhiguo</namePart>
<namePart type="family">Yang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jiacheng</namePart>
<namePart type="family">Pi</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dongsheng</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ruizhi</namePart>
<namePart type="family">Zhong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Peipei</namePart>
<namePart type="family">Xu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Wenjie</namePart>
<namePart type="family">Ruan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<originInfo>
<dateIssued>2026-10</dateIssued>
</originInfo>
<typeOfResource>text</typeOfResource>
<relatedItem type="host">
<titleInfo>
<title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
</titleInfo>
<name type="personal">
<namePart type="given">Hung-Yi</namePart>
<namePart type="family">Lee</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Sunipa</namePart>
<namePart type="family">Dev</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Jessy</namePart>
<namePart type="family">Li</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Desmond</namePart>
<namePart type="family">Elliott</namePart>
<role>
<roleTerm authority="marcrelator" type="text">editor</roleTerm>
</role>
</name>
<originInfo>
<publisher>Association for Computational Linguistics</publisher>
<place>
<placeTerm type="text">Budapest, Hungary</placeTerm>
</place>
</originInfo>
<genre authority="marcgt">conference publication</genre>
<identifier type="isbn">979-8-89176-459-0</identifier>
</relatedItem>
<abstract>Multimodal Large Language Models (MLLMs) have achieved impressive progress in vision-language understanding, yet they remain susceptible to object hallucination, which refers to generating descriptions of objects that do not exist in the input image. Through comprehensive analysis, we uncover a fundamental trade-off in MLLMs: strengthening visual attention enhances grounding completeness but also increases hallucinations, especially additive hallucinations that appear in the trailing part of the generated sentence. To alleviate this trade-off, we propose **GLVAC**, **G**lobal-to-**L**ocal **V**isual **A**ttention **C**orrection, a training-free dual-path inference framework that adaptively combines global conservatism and local enhancement in visual attention. The global path maintains stable and grounded predictions under moderate visual attention, while the local correction branch is dynamically activated for high-uncertainty tokens identified via entropy-based estimation. A confidence-guided gating mechanism then determines whether to adopt the correction output, ensuring safe and reliable refinement. GLVAC can be seamlessly integrated into existing MLLMs without retraining or external modules, effectively reducing object hallucinations while maintaining perceptual completeness across diverse benchmarks and model families. Our code is available on https://github.com/Trusted-LLM/GLVAC.</abstract>
<identifier type="citekey">huang-etal-2026-mitigating-object</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.2272/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>51671</start>
<end>51689</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Mitigating Object Hallucination in MLLMs with Global-to-Local Visual Attention Correction
%A Huang, Xingxing
%A Yang, Zhiguo
%A Pi, Jiacheng
%A Xu, Dongsheng
%A Zhong, Ruizhi
%A Xu, Peipei
%A Ruan, Wenjie
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F huang-etal-2026-mitigating-object
%X Multimodal Large Language Models (MLLMs) have achieved impressive progress in vision-language understanding, yet they remain susceptible to object hallucination, which refers to generating descriptions of objects that do not exist in the input image. Through comprehensive analysis, we uncover a fundamental trade-off in MLLMs: strengthening visual attention enhances grounding completeness but also increases hallucinations, especially additive hallucinations that appear in the trailing part of the generated sentence. To alleviate this trade-off, we propose **GLVAC**, **G**lobal-to-**L**ocal **V**isual **A**ttention **C**orrection, a training-free dual-path inference framework that adaptively combines global conservatism and local enhancement in visual attention. The global path maintains stable and grounded predictions under moderate visual attention, while the local correction branch is dynamically activated for high-uncertainty tokens identified via entropy-based estimation. A confidence-guided gating mechanism then determines whether to adopt the correction output, ensuring safe and reliable refinement. GLVAC can be seamlessly integrated into existing MLLMs without retraining or external modules, effectively reducing object hallucinations while maintaining perceptual completeness across diverse benchmarks and model families. Our code is available on https://github.com/Trusted-LLM/GLVAC.
%U https://aclanthology.org/2026.emnlp-main.2272/
%P 51671-51689
Markdown (Informal)
[Mitigating Object Hallucination in MLLMs with Global-to-Local Visual Attention Correction](https://aclanthology.org/2026.emnlp-main.2272/) (Huang et al., EMNLP 2026)
ACL