AI Paper Atlas

Research map / Video / Video-language understanding

Linear Scaling Video VLMs for Long Video Understanding

Cristobal Eyzaguirre, Jiajun Wu, Juan Carlos Niebles

ECCV 2026 · Video › Video-language understanding · approach: Long video understanding · 0 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 importance-based StateKV state instead of ReKV sliding window (compute-matched)StateKV · VideoMME, MLVU, OVOBenchaccuracy: about 10 points on averageImproved
use StateKV (B=4096) instead of full self-attentionStateKV · VideoMME, MLVU, OVOBenchaccuracy: within roughly a point on averageNo clear effect
run InternVL3-8B with StateKV instead of InternVL3-1B with full SA at similar computeStateKV · VideoMME (no subtitles, 512 frames)accuracy: 46.2 -> 62.5Improved
increase StateKV cache budget B (ablation)StateKV · VideoMME short videos (InternVL3)accuracy: 75.0Improved
replace full SA with ReKV (R=16) on InternVL3-2BReKV · MLVU (512 frames)accuracy: 56.6104 -> 5.4931Hurt
use StateKV (B=4096) instead of full SA on InternVL3-2BStateKV · MLVU (512 frames)accuracy: 56.6104 -> 57.3099No clear effect
use StateKV (B=4096) instead of full SA on InternVL3-8BStateKV · MLVU (512 frames)accuracy: 61.14 -> 62.85No clear effect
use StateKV (B=4096) instead of full SA on Qwen3-VL-8BStateKV · MLVU (512 frames)accuracy: 75.82 -> 71.38Hurt
use ReKV (R=16) instead of full SA on Qwen3-VL-8BReKV · VideoMME (no subtitles, 512 frames)accuracy: 70.52 -> 55.52Hurt
select tokens by attention instead of keeping recent frames (256 tokens vs 1 frame) (ablation)StateKV · Attention mass analysis (InternVL3-1B/2B/8B)fraction of historical attention mass: gap about 0.59/0.57/0.62Improved
raise cache budget B from 256 to 4096 tokens (ablation)StateKV · Attention mass analysis (InternVL3-1B)historical attention mass captured: 0.71 -> 0.93Improved
use Triton fused score kernel instead of eager attention for cache building (ablation)StateKV · wall time per frame (L40S)wall timeImproved

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