AI Paper Atlas

Research map / Video / Video-language understanding

VideoNSA: Native Sparse Attention Scales Video Understanding

Enxin Song, Wenhao Chai, Shusheng Yang, Ethan J. Armand, Xiaojun Shan, Haiyang Xu, Jianwen Xie, Zhuowen Tu

ICLR 2026 · Video › Video-language understanding · approach: Long video understanding · 11 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use learnable native sparse attention (VideoNSA) instead of dense attentionQwen2.5-VL-7B · LongTimeScopeaccuracy: 40.7 -> 44.4Improved
use VideoNSAQwen2.5-VL-7B · Tomatoaccuracy: 22.6 -> 26.5Improved
use VideoNSAQwen2.5-VL-7B · VSIBenchaccuracy: 29.7 -> 36.1Improved
use VideoNSAQwen2.5-VL-7B · LongVideoBenchaccuracy: 58.7 -> 60.0Mixed
replace token compression with sparse attentionQwen2.5-VL-7B · long video / VSIBenchaccuracy: FastV 41.8 vs VideoNSA 51.8 on MLVUImproved
use all three branches (CMP, SLC, SWD) instead of one (ablation)VideoNSA · LongVideoBenchaccuracy: 49.1 -> 60.0Improved
use all three branches instead of selection only (ablation)VideoNSA · TimeScopeaccuracy: 63.7 -> 83.7Improved
transfer learned sparse weights to dense attention (Dense-NSA) (ablation)Qwen2.5-VL-7B · VSIBenchaccuracy: 30.6 -> 33.1Improved
transfer sparse weights to dense attention (ablation)Qwen2.5-VL-7B · LongVideoBenchaccuracy: 57.8 -> 56.1Hurt
keep runtime sparsity and dynamic gating vs dense transfer (ablation)VideoNSA · LongTimeScopeaccuracy: 40.9 -> 44.4Improved
increase global attention (more blocks) vs local sliding window at fixed budget (ablation)VideoNSA · video benchmarksaccuracyImproved
scale context beyond training length (36K to 128K) (ablation)VideoNSA · video benchmarksaccuracyImproved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy