AI Paper Atlas

Research map / Video / Video-language understanding

State-Space Hierarchical Compression with Gated Attention and Learnable Sampling for Hour-Long Video Understanding in Large Multimodal Models

Geewook Kim, Minjoon Seo

AAAI 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
replace average-pooling queries with learnable gated patch aggregation (GPA) (ablation)BIMBA-style bi-Mamba compressor · LVBench/MLVU/VideoMME averageaverage accuracy: 45.44 -> 52.22Improved
add time-axis aggregator (TAA) on top of GPA (ablation)MambaMia · LVBench/MLVU/VideoMME averageaverage accuracy: 52.22 -> 53.56Improved
use Mamba V1 instead of V2 (ablation)MambaMia · LVBench/MLVU/VideoMME averageaverage accuracy: 53.56 -> 53.13No clear effect
use bi-directional instead of uni-directional compressor blocks (ablation)MambaMia · LVBench/MLVU/VideoMME averageaverage accuracy: 51.86 -> 53.56Improved
use delta-based frame sampling instead of uniform 50% sampling (ablation)MambaMia final model · LVBench/MLVU/VideoMME averageaverage accuracy: 56.30 -> 56.95Improved
use delta-based frame sampling instead of uniform 50% sampling (ablation)MambaMia final model · LVBenchaccuracy: 43.38 -> 44.61Improved
add delta-based secondary sampling (halving tokens) versus none (ablation)MambaMia final model · LVBench/MLVU/VideoMME averageaverage accuracy: 57.27 -> 56.95No clear effect
replace vanilla MLP projector with proposed compressionPythia-2.8B LMM · LVBench/MLVU/VideoMME averageaverage accuracy: 32.78 -> 41.71Improved
replace vanilla MLP projector with proposed compressionMamba2-2.7B LMM · LVBench/MLVU/VideoMME averageaverage accuracy: 34.14 -> 39.85Improved
use a Mamba2 LLM instead of Pythia with vanilla projectorVanilla LMM · LVBench/MLVU/VideoMME averageaverage accuracy: 32.78 -> 34.14No clear effect

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 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