AI Paper Atlas

Research map / Video / Video-language understanding

MeToM: Metadata-Guided Token Merging for Efficient Video LLMs

Zhuojie Wu, Shijie Wang, Xin Yu

CVPR 2026 · Video › Video-language understanding · approach: Long video understanding

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
use codec-metadata-guided token merging (MeToM) instead of uniform pruning, 50% budgetLLaVA-OneVision-7B · 5 video QA benchmarks averagerelative accuracy: 102.0 vs 101.2 (ToMe), 100.4 (HoliTom)Improved
use MeToM at an aggressive 30% token budgetLLaVA-OneVision-7B · 5 video QA benchmarks averagerelative accuracy: 101.2 vs 98.4 (FastV), 99.6 (HoliTom)Improved
apply FastV at 30% budgetLLaVA-OneVision-7B · 5 video QA benchmarks averagerelative accuracy: 98.4Hurt
use MeToM on LLaVA-Video-7B at 30% budgetLLaVA-Video-7B · 5 video QA benchmarks averagerelative accuracy: 98.4 vs 96.0 (FastV), 97.4 (HoliTom)Improved
use MeToM on Qwen2VL-7B at 30% budgetQwen2VL-7B · VideoMME, LongVideoBenchrelative accuracy: 101.3 vs 97.9 (ToMe)Improved
use MeToM on LLaVA-Video-72B at 50% budgetLLaVA-Video-72B · VideoMMErelative accuracy: 101.4Improved
use MeToM on LLaVA-Video-72B at 30% budgetLLaVA-Video-72B · VideoMMErelative accuracy: 99.1 vs 97.1 (ToMe), 98.3 (DyCoke*)Improved
add residual-guided patch merging before the vision encoderMeToM · Vision tower latencyvision tower time: 525ms -> 450msImproved
use MeToM overall for inference speedLLaVA-OneVision-7B · TTFTspeedup: 2.65x speedup; LLM backbone 1329ms -> 215msImproved
use codec-metadata preprocessing instead of similarity-based preprocessingtoken compression methods · pre-processing overheadpre-processing time: 53ms vs HoliTom 92ms, DyCoke 58ms (baseline 50ms)Improved

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