Infoblock alpha
cs-cv/vision-language IB-arxiv-2607.28463-mlvu-gain measured

VisualRouter improved Qwen2.5-VL-7B by 11.6% on MLVU over uniform sampling.

Active Evidence state is recorded explicitly and may be revised by a later event.

Observed quantities

gain_pct 11.6

01

Primary source

  1. arxiv · 2607.28463 · v1 · abstract
Title
VisualRouter: Query-Grounded Visual Sampling for Long Video Understanding
Authors
Haiyue Zhang, Yi Bin, Xun Jiang, Zeyu Ma, Duo Peng, Guoqing Wang, Yang Yang, Heng Tao Shen
Version
v1
Check
Support quote found word for word in the arXiv abstract v1, 2026-10-08
Supporting quote from the source
VisualRouter consistently improves multiple LVLMs over uniform sampling, achieving gains of 5.2%, 7.7%, and 11.6% on Video-MME, LongVideoBench, and MLVU with Qwen2.5-VL-7B

02

Boundary of this block

No additional public limitation was recorded.

03

Connections

Outgoing

No outgoing public edges.

Incoming

No incoming public edges.