AI Paper Atlas

Research map / Video / Video-language understanding

SVAgent: Storyline-guided Long Video Understanding via Cross-Modal Multi-Agent Collaboration

Zhongyu Yang, Zuhao Yang, Shuo Zhan, Tan Yue, Wei Pang, Yingfang Yuan

CVPR 2026 · Video › Video-language understanding · approach: Video question answering · 5 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
add SVAgent multi-agent frameworkQwen2.5-VL-3B · LongVideoBenchaccuracy: 53.0 -> 59.7Improved
add SVAgentQwen3-VL-4B · MLVUaccuracy: 53.4 -> 64.9Improved
add SVAgentQwen2.5-VL-7B · VideoMME mediumaccuracy: 69.0 -> 75.8Improved
add SVAgentQwen2.5-VL-7B · VideoMME longaccuracy: 52.3 -> 57.3Improved
add storyline agent alone (ablation)Qwen2.5-VL-7B · LongVideoBench and LVBenchaccuracy: +5.6 LongVideoBench, +4.9 LVBenchImproved
add textual verification (ablation)Qwen2.5-VL-7B · MLVUaccuracy: +3.8Improved
add visual verification (ablation)Qwen2.5-VL-7B · VideoMME longaccuracy: +3.5Improved
add meta-decision agent (ablation)Qwen2.5-VL-7B · LongVideoBenchaccuracy: +3.0Improved
swap SigLIP for CLIP or LongCLIP as DPP retrieval encoder (ablation)SVAgent · LongVideoBenchaccuracy: 60.7 -> 60.2 (CLIP), 59.9 (LongCLIP)No clear effect
raise intersection ratio tau (ablation)SVAgent · Video-MMEaccuracy: 64.1 -> 67.2; latency 11.8s -> 38.5sMixed
increase refinement iterations T up to 10 (ablation)SVAgent · Video-MMEaccuracy: accuracy up to 68.2; latency 24.9s -> 57.3sMixed
use SVAgent frame selection instead of uniform sampling at 8 framesQwen2.5-VL-7B · Video-MMEaccuracy: uniform 59.8% at 32 frames vs SVAgent 60.7% at 8 framesImproved

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