AI Paper Atlas

Research map / Video / Video-language understanding

FluxMem: Adaptive Hierarchical Memory for Streaming Video Understanding

Yiweng Xie, Bo He, Junke Wang, Xiangyu Zheng, Ziyi Ye, Zuxuan Wu

CVPR 2026 · Video › Video-language understanding · approach: Long video understanding · 26 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
apply FluxMem hierarchical memoryQwen2.5-VL-7B · StreamingBench real-timeaccuracy: 73.9 -> 76.4Improved
apply FluxMem hierarchical memoryQwen2.5-VL-7B · OVO-Bench overallaccuracy: 49.8 -> 53.3Improved
apply FluxMem hierarchical memoryQwen2.5-VL-7B · MLVUM-Avg accuracy: 67.9 -> 73.1Improved
apply FluxMem hierarchical memoryQwen2.5-VL-7B · VideoMME Longaccuracy: 53.8 -> 54.0No clear effect
apply FluxMem hierarchical memoryQwen2.5-VL-7B · MLVU / OVO-Benchlatency: 3614 -> 2014 ms; 2701 -> 812 msImproved
add mid-term memory (TAS) to S+L (ablation)FluxMem · MLVUM-Avg accuracy: 70.9 -> 73.1Improved
add short-term memory to M+L (ablation)FluxMem · StreamingBench real-timeaccuracy: 75.8 -> 76.4Improved
add long-term memory (SDC) to S+M (ablation)FluxMem · Average scoreaverage: 70.2 -> 71.6Improved
add short-term memory to long-term only (ablation)FluxMem · StreamingBench real-timeaccuracy: 75.9 -> 77.0Improved
add short-term memory to long-term only (ablation)FluxMem · VideoMME w/o subaccuracy: 62.0 -> 61.9Hurt
use adaptive Otsu threshold instead of best fixed (mid-term) (ablation)FluxMem · MLVUdrop ratio at equal accuracy: 65.5 at 29.4% -> 65.6 at 42.8% dropImproved
use adaptive Otsu threshold instead of best fixed (long-term) (ablation)FluxMem · MLVUdrop ratio at equal accuracy: 65.6 at 63.8% -> 65.4 at 84.5% dropImproved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 0 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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