@inproceedings{zhu-etal-2026-visa,
title = "{VISA}: Video Skeleton-Aware Efficient Frame Selection for Long Video Understanding",
author = "Zhu, Qihui and
Wen, Zijian and
Zhang, Tao and
Zhang, Mengjie and
Ye, Shenghao and
Jin, Dong and
Chen, Shuangwu and
Tan, Xiaobin and
Zheng, Quan and
Jianyang",
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.1142/",
pages = "26175--26188",
ISBN = "979-8-89176-459-0",
abstract = "Long-video understanding remains challenging for multimodal large language models (MLLMs), as informative visual evidence is often sparsely distributed across many redundant frames. Although sampling more frames can improve coverage, it also increases visual encoding overhead and consumes more visual tokens, making compact frame selection essential. However, existing methods often sample candidate frames uniformly before selection, which may require decoding most frames of the video. Moreover, simply selecting frames by query relevance may overlook the global temporal structure needed for coherent video understanding. To address these limitations, we propose \textbf{VISA}, a \textbf{VI}deo \textbf{S}keleton-\textbf{A}ware framework for efficient long-video frame selection. VISA exploits sparse I-frame anchors from the native GOP structure to avoid dense candidate decoding and constructs a video skeleton that preserves the coarse temporal progression. Query relevance is aggregated over this skeleton to select informative anchors, followed by cross-GOP refinement to obtain final frames. Extensive experiments show that VISA consistently improves long-video understanding, achieving up to 8.8{\%} relative gain over uniform sampling. Meanwhile, VISA reduces average end-to-end latency by 32.9{\%}{--}53.6{\%} compared with recent frame selection baselines on LongVideoBench, demonstrating both effectiveness and efficiency for long-video frame selection."
}<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="zhu-etal-2026-visa">
<titleInfo>
<title>VISA: Video Skeleton-Aware Efficient Frame Selection for Long Video Understanding</title>
</titleInfo>
<name type="personal">
<namePart type="given">Qihui</namePart>
<namePart type="family">Zhu</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Zijian</namePart>
<namePart type="family">Wen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Tao</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Mengjie</namePart>
<namePart type="family">Zhang</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shenghao</namePart>
<namePart type="family">Ye</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Dong</namePart>
<namePart type="family">Jin</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Shuangwu</namePart>
<namePart type="family">Chen</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Xiaobin</namePart>
<namePart type="family">Tan</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name type="personal">
<namePart type="given">Quan</namePart>
<namePart type="family">Zheng</namePart>
<role>
<roleTerm authority="marcrelator" type="text">author</roleTerm>
</role>
</name>
<name>
<namePart>Jianyang</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>Long-video understanding remains challenging for multimodal large language models (MLLMs), as informative visual evidence is often sparsely distributed across many redundant frames. Although sampling more frames can improve coverage, it also increases visual encoding overhead and consumes more visual tokens, making compact frame selection essential. However, existing methods often sample candidate frames uniformly before selection, which may require decoding most frames of the video. Moreover, simply selecting frames by query relevance may overlook the global temporal structure needed for coherent video understanding. To address these limitations, we propose VISA, a VIdeo Skeleton-Aware framework for efficient long-video frame selection. VISA exploits sparse I-frame anchors from the native GOP structure to avoid dense candidate decoding and constructs a video skeleton that preserves the coarse temporal progression. Query relevance is aggregated over this skeleton to select informative anchors, followed by cross-GOP refinement to obtain final frames. Extensive experiments show that VISA consistently improves long-video understanding, achieving up to 8.8% relative gain over uniform sampling. Meanwhile, VISA reduces average end-to-end latency by 32.9%–53.6% compared with recent frame selection baselines on LongVideoBench, demonstrating both effectiveness and efficiency for long-video frame selection.</abstract>
<identifier type="citekey">zhu-etal-2026-visa</identifier>
<location>
<url>https://aclanthology.org/2026.emnlp-main.1142/</url>
</location>
<part>
<date>2026-10</date>
<extent unit="page">
<start>26175</start>
<end>26188</end>
</extent>
</part>
</mods>
</modsCollection>
%0 Conference Proceedings
%T VISA: Video Skeleton-Aware Efficient Frame Selection for Long Video Understanding
%A Zhu, Qihui
%A Wen, Zijian
%A Zhang, Tao
%A Zhang, Mengjie
%A Ye, Shenghao
%A Jin, Dong
%A Chen, Shuangwu
%A Tan, Xiaobin
%A Zheng, Quan
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%A Jianyang
%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 zhu-etal-2026-visa
%X Long-video understanding remains challenging for multimodal large language models (MLLMs), as informative visual evidence is often sparsely distributed across many redundant frames. Although sampling more frames can improve coverage, it also increases visual encoding overhead and consumes more visual tokens, making compact frame selection essential. However, existing methods often sample candidate frames uniformly before selection, which may require decoding most frames of the video. Moreover, simply selecting frames by query relevance may overlook the global temporal structure needed for coherent video understanding. To address these limitations, we propose VISA, a VIdeo Skeleton-Aware framework for efficient long-video frame selection. VISA exploits sparse I-frame anchors from the native GOP structure to avoid dense candidate decoding and constructs a video skeleton that preserves the coarse temporal progression. Query relevance is aggregated over this skeleton to select informative anchors, followed by cross-GOP refinement to obtain final frames. Extensive experiments show that VISA consistently improves long-video understanding, achieving up to 8.8% relative gain over uniform sampling. Meanwhile, VISA reduces average end-to-end latency by 32.9%–53.6% compared with recent frame selection baselines on LongVideoBench, demonstrating both effectiveness and efficiency for long-video frame selection.
%U https://aclanthology.org/2026.emnlp-main.1142/
%P 26175-26188
Markdown (Informal)
[VISA: Video Skeleton-Aware Efficient Frame Selection for Long Video Understanding](https://aclanthology.org/2026.emnlp-main.1142/) (Zhu et al., EMNLP 2026)
ACL
- Qihui Zhu, Zijian Wen, Tao Zhang, Mengjie Zhang, Shenghao Ye, Dong Jin, Shuangwu Chen, Xiaobin Tan, Quan Zheng, and Jianyang. 2026. VISA: Video Skeleton-Aware Efficient Frame Selection for Long Video Understanding. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 26175–26188, Budapest, Hungary. Association for Computational Linguistics.