X X -Stream: Exploring MLLMs as Multiplexers for Multi-Stream Understanding

Peiwen Sun *, Xudong Lu *, Huadai Liu *, Yang Bo, Dongming Wu, Huankang Guan, Minghong Cai, Jinpeng Chen, Xintong Guo, Shuhan Li, Fang Liu, Rui Liu, Xiangyu Yue · 2026

Read X X -Stream: Exploring MLLMs as Multiplexers for Multi-Stream Understanding by Peiwen Sun *, Xudong Lu *, Huadai Liu *, Yang Bo, Dongming Wu, Huankang Guan, Minghong Cai, Jinpeng Chen, Xintong Guo, Shuhan Li, Fang Liu, Rui Liu, Xiangyu Yue (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2606-02482.txt.

Description

Sun and co-authors build the first benchmark for multi-stream video understanding. Real uses demand many streams at once. Live sports, driving, and multi-screen work all mix feeds. Past benchmarks test one stream at a time. They leave cross-stream reasoning unmeasured.

Their benchmark, X-Stream, holds 4,220 question-answer pairs across 932 videos. Eleven subtasks span multi-window, multi-view, and multi-device settings. A dual-verification pipeline stops the data from leaning on one stream. The authors also frame models as naive multiplexers, using signal multiplexing theory. Online tests show top models struggle. Scores reach only about 50 percent. Proactive ability is poor. X-Stream exposes the trade-off in current multiplexing schemes. It gives both a test protocol and guidance for next designs.

Catalog record

CREATORPeiwen Sun *, Xudong Lu *, Huadai Liu *, Yang Bo, Dongming Wu, Huankang Guan, Minghong Cai, Jinpeng Chen, Xintong Guo, Shuhan Li, Fang Liu, Rui Liu, Xiangyu Yue
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE70 KB
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2606.02482

Browse the full library · About the Tower of Babel