@inproceedings{jeong-etal-2026-keeping,
title = "Keeping Eyes on the Image: Dynamic Visual Attention Recalibration for Vision-Language Reasoning",
author = "Jeong, Yuna and
Park, Junhyeong and
Kang, Mingyu and
Seo, Hyein and
Kim, Keon and
Choi, Yong Suk",
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.147/",
pages = "3398--3418",
ISBN = "979-8-89176-459-0",
abstract = "Large Vision-Language Models (LVLMs) achieve strong performance on various vision-language tasks, yet Chain-of-Thought (CoT) generation often weakens attention to visual tokens, degrading evidence-grounded reasoning. During the later stages of autoregressive decoding, models tend to increasingly rely on previously generated text rather than visual evidence, leading to plausible but visually inconsistent rationales. We propose DVAR (Dynamic Visual Attention Recalibration), a training-free method that selectively amplifies visually relevant tokens to mitigate visual attention decay during decoding. DVAR operates at three levels: (i) prompt level structure to encourage visually grounded reasoning; (ii) embedding level relevance weights derived from image{--}text alignment; and (iii) decoding level scheduling that adjusts recalibration by tracking shifts in visual attention. Without sensitive model-specific hyperparameter retuning, DVAR yields consistent performance improvements across six benchmarks on LLaVA and Qwen-VL models, and better preserves visual attention throughout decoding."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="jeong-etal-2026-keeping">
<titleInfo>
<title>Keeping Eyes on the Image: Dynamic Visual Attention Recalibration for Vision-Language Reasoning</title>
</titleInfo>
<name type="personal">
<namePart type="given">Yuna</namePart>
<namePart type="family">Jeong</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Junhyeong</namePart>
<namePart type="family">Park</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mingyu</namePart>
<namePart type="family">Kang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Hyein</namePart>
<namePart type="family">Seo</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Keon</namePart>
<namePart type="family">Kim</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Yong</namePart>
<namePart type="given">Suk</namePart>
<namePart type="family">Choi</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>Large Vision-Language Models (LVLMs) achieve strong performance on various vision-language tasks, yet Chain-of-Thought (CoT) generation often weakens attention to visual tokens, degrading evidence-grounded reasoning. During the later stages of autoregressive decoding, models tend to increasingly rely on previously generated text rather than visual evidence, leading to plausible but visually inconsistent rationales. We propose DVAR (Dynamic Visual Attention Recalibration), a training-free method that selectively amplifies visually relevant tokens to mitigate visual attention decay during decoding. DVAR operates at three levels: (i) prompt level structure to encourage visually grounded reasoning; (ii) embedding level relevance weights derived from image–text alignment; and (iii) decoding level scheduling that adjusts recalibration by tracking shifts in visual attention. Without sensitive model-specific hyperparameter retuning, DVAR yields consistent performance improvements across six benchmarks on LLaVA and Qwen-VL models, and better preserves visual attention throughout decoding.</abstract>
<identifier type="citekey">jeong-etal-2026-keeping</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.147/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>3398</start>
<end>3418</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T Keeping Eyes on the Image: Dynamic Visual Attention Recalibration for Vision-Language Reasoning
%A Jeong, Yuna
%A Park, Junhyeong
%A Kang, Mingyu
%A Seo, Hyein
%A Kim, Keon
%A Choi, Yong Suk
%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 jeong-etal-2026-keeping
%X Large Vision-Language Models (LVLMs) achieve strong performance on various vision-language tasks, yet Chain-of-Thought (CoT) generation often weakens attention to visual tokens, degrading evidence-grounded reasoning. During the later stages of autoregressive decoding, models tend to increasingly rely on previously generated text rather than visual evidence, leading to plausible but visually inconsistent rationales. We propose DVAR (Dynamic Visual Attention Recalibration), a training-free method that selectively amplifies visually relevant tokens to mitigate visual attention decay during decoding. DVAR operates at three levels: (i) prompt level structure to encourage visually grounded reasoning; (ii) embedding level relevance weights derived from image–text alignment; and (iii) decoding level scheduling that adjusts recalibration by tracking shifts in visual attention. Without sensitive model-specific hyperparameter retuning, DVAR yields consistent performance improvements across six benchmarks on LLaVA and Qwen-VL models, and better preserves visual attention throughout decoding.
%U https://aclanthology.org/2026.emnlp-main.147/
%P 3398-3418
Markdown (Informal)
[Keeping Eyes on the Image: Dynamic Visual Attention Recalibration for Vision-Language Reasoning](https://aclanthology.org/2026.emnlp-main.147/) (Jeong et al., EMNLP 2026)
ACL