@inproceedings{liang-etal-2026-vdahm,
title = "{VDAHM}: Vision-Delimiter-aware Attention Head Modulation for Enhancing Multi-Image Understanding in Large Vision-Language Models",
author = "Liang, Zhibiao and
Cheng, Junyu and
Zeng, Ziqian",
editor = "Lee, Hung-Yi and
Dev, Sunipa and
Li, Jessy and
Elliott, Desmond",
booktitle = "Findings of the {A}ssociation for {C}omputational {L}inguistics: {EMNLP} 2026",
month = oct,
year = "2026",
address = "Budapest, Hungary",
publisher = "Association for Computational Linguistics",
url = "https://aclanthology.org/2026.findings-emnlp.125/",
pages = "2542--2561",
ISBN = "979-8-89176-462-0",
abstract = "Although Large Vision-Language Models (LVLMs) excel at single-image understanding tasks, their performance often degrades significantly when processing multi-image inputs due to unintended cross-image information leakage, in which visual content from different images is erroneously fused and image-specific contextual information is lost. Current approaches attempt to mitigate this by inserting special vision delimiter tokens between images during input formatting to help models distinguish image boundaries and preserve context. However, our analysis reveals that such delimiters alone yield limited gains because models do not sufficiently perceive or respond to these boundary signals. To address this, we propose Vision-Delimiter-aware Attention Head Modulation (VDAHM), a training-free method that modulates the output of delimiter-sensitive attention heads, reorienting the model{'}s internal attention flow to strengthen its structural perception of boundary signals. Extensive evaluations across multiple benchmarks, including Mantis, BLINK, MuirBench, and Qbench2, demonstrate that VDAHM consistently improves LVLMs performance with virtually no additional computational cost. The source code is available at [https://github.com/ZeroNLP/VDAHM](https://github.com/ZeroNLP/VDAHM)."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="liang-etal-2026-vdahm">
<titleInfo>
<title>VDAHM: Vision-Delimiter-aware Attention Head Modulation for Enhancing Multi-Image Understanding in Large Vision-Language Models</title>
</titleInfo>
<name type="personal">
<namePart type="given">Zhibiao</namePart>
<namePart type="family">Liang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Junyu</namePart>
<namePart type="family">Cheng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Ziqian</namePart>
<namePart type="family">Zeng</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>Findings of the Association for Computational Linguistics: EMNLP 2026</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">979-8-89176-462-0</identifier>
</relatedItem>
<abstract>Although Large Vision-Language Models (LVLMs) excel at single-image understanding tasks, their performance often degrades significantly when processing multi-image inputs due to unintended cross-image information leakage, in which visual content from different images is erroneously fused and image-specific contextual information is lost. Current approaches attempt to mitigate this by inserting special vision delimiter tokens between images during input formatting to help models distinguish image boundaries and preserve context. However, our analysis reveals that such delimiters alone yield limited gains because models do not sufficiently perceive or respond to these boundary signals. To address this, we propose Vision-Delimiter-aware Attention Head Modulation (VDAHM), a training-free method that modulates the output of delimiter-sensitive attention heads, reorienting the model’s internal attention flow to strengthen its structural perception of boundary signals. Extensive evaluations across multiple benchmarks, including Mantis, BLINK, MuirBench, and Qbench2, demonstrate that VDAHM consistently improves LVLMs performance with virtually no additional computational cost. The source code is available at [https://github.com/ZeroNLP/VDAHM](https://github.com/ZeroNLP/VDAHM).</abstract>
<identifier type="citekey">liang-etal-2026-vdahm</identifier>
<location>
<url>https://aclanthology.org/2026.findings-emnlp.125/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>2542</start>
<end>2561</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T VDAHM: Vision-Delimiter-aware Attention Head Modulation for Enhancing Multi-Image Understanding in Large Vision-Language Models
%A Liang, Zhibiao
%A Cheng, Junyu
%A Zeng, Ziqian
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Findings of the Association for Computational Linguistics: EMNLP 2026
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-462-0
%F liang-etal-2026-vdahm
%X Although Large Vision-Language Models (LVLMs) excel at single-image understanding tasks, their performance often degrades significantly when processing multi-image inputs due to unintended cross-image information leakage, in which visual content from different images is erroneously fused and image-specific contextual information is lost. Current approaches attempt to mitigate this by inserting special vision delimiter tokens between images during input formatting to help models distinguish image boundaries and preserve context. However, our analysis reveals that such delimiters alone yield limited gains because models do not sufficiently perceive or respond to these boundary signals. To address this, we propose Vision-Delimiter-aware Attention Head Modulation (VDAHM), a training-free method that modulates the output of delimiter-sensitive attention heads, reorienting the model’s internal attention flow to strengthen its structural perception of boundary signals. Extensive evaluations across multiple benchmarks, including Mantis, BLINK, MuirBench, and Qbench2, demonstrate that VDAHM consistently improves LVLMs performance with virtually no additional computational cost. The source code is available at [https://github.com/ZeroNLP/VDAHM](https://github.com/ZeroNLP/VDAHM).
%U https://aclanthology.org/2026.findings-emnlp.125/
%P 2542-2561
Markdown (Informal)
[VDAHM: Vision-Delimiter-aware Attention Head Modulation for Enhancing Multi-Image Understanding in Large Vision-Language Models](https://aclanthology.org/2026.findings-emnlp.125/) (Liang et al., Findings 2026)
ACL