AI Paper Atlas

Research map / Video / Video-language understanding

ProtoKV: Streaming Video Understanding under Delayed Query with Summary-State Memory

Le Tu Ngoc Minh, Jinyeong Lim, Dongsu Han

ICML 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
summarize far history as prototype bank instead of retaining tokens (ProtoKV vs InfiniPot-V)LLaVA-OV-7B · RVS-Egoaccuracy (%): 57.8 -> 58.6Improved
ProtoKV vs InfiniPot-V on StreamingBench real-timeQwen2.5-VL-7B · StreamingBenchaccuracy (%): 76.4 -> 77.3Improved
add far prototype memory to near window (ablation)ProtoKV · RVS-Ego at delay 30 minaccuracy (%): 40.5 -> 54.3Improved
add exact near window to far-only memory (ablation)ProtoKV · RVS-Ego at delay 0accuracy (%): 54.5 -> 59.0Improved
add residual statistics (ablation)ProtoKV · RVS-Ego at delay 30 minaccuracy (%): 51.0 -> 54.3Improved
add mass-aware log bias (ablation)ProtoKV · RVS-Ego at delay 30 minaccuracy (%): 51.0 -> 54.3Improved
add continuity-aware assignment (spatial term) (ablation)ProtoKV · RVS-Ego at delay 30 minaccuracy (%): 47.5 -> 54.3Improved
add residual statistics at no delay (ablation)ProtoKV · RVS-Ego at delay 0accuracy (%): 58.5 -> 59.0No clear effect
ProtoKV vs SWA/InfiniPot-V at long query delayProtoKV · RVS-Movie, largest delaysaccuracy gain (pp): +20.4pp over SWA, +12.5pp over token retentionImproved
ProtoKV vs InfiniPot-V on offline MLVU at 3K budgetQwen2-VL-7B · MLVUaccuracy (%): 67.2 -> 67.0No clear effect
ProtoKV vs InfiniPot-V on offline VideoMME at 3K budgetQwen2-VL-7B · VideoMMEaccuracy (%): 61.2 -> 62.1Improved
replace ReKV offloading with GPU-resident ProtoKVLLaVA-OV-7B · RVS-EgoTTFT (s): 1.43 -> 1.06Improved

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