AI Paper Atlas

Research map / Video / Video-language understanding

Multi-Granular Spatio-Temporal Token Merging for Training-Free Acceleration of Video LLMs

Jeongseok Hyun, Sukjun Hwang, Su Ho Han, Taeoh Kim, Inwoong Lee, Dongyoon Wee, Joon-Young Lee, Seon Joo Kim, Minho Shim

ICCV 2025 · Video › Video-language understanding · approach: Video LLMs with fine-grained understanding · 22 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
use STTM instead of other query-agnostic merging (ToMe, DyCoke-stage1)LLaVA-Video-7B, 30% token budget · VNBenchrelative accuracy: 82.7 -> 98.0 (vs ToMe)Improved
use multi-granular quadtree spatial merging instead of single-granularity merging (ablation)STTM spatial merging · VNBenchrelative accuracy: 80.0 -> 98.7Improved
single-granularity merging on VideoMME (ablation)STTM spatial merging · VideoMMErelative accuracy: 98.4 vs 99.2No clear effect
use 4x4 root nodes instead of 2x2 in quadtree (ablation)STTM spatial merging · VNBenchrelative accuracy: 73.1 -> 98.7Improved
add temporal merging after spatial merging (sequential decomposition) (ablation)STTM · VNBench / VideoMMErelative token count: R.NV 57.2 -> 25.8 on VNBench with accuracy 98.7 -> 98.0Improved
joint octree spatio-temporal merging instead of decomposed merging (ablation)STTM · VNBenchrelative accuracy: 98.0 -> 73.3Hurt
use top-left approximate merge destination instead of most similar (ablation)STTM temporal merging · 30% budget averagerelative accuracy: 97.7 -> 98.8; TTFT 39.6 -> 31.4No clear effect
reassign position ids instead of merging or keeping RoPE (ablation)STTM positional embedding · VNBenchrelative accuracy: 96.0 -> 98.0Improved
merge tokens at later LLM layer (19 instead of 3) (ablation)STTM · VNBenchrelative accuracy vs TTFT: accuracy 98.0 -> 99.5; relative TTFT 30.9 -> 75.5Mixed
remove special grid token in LLaVA-Video (ablation)LLaVA-Video-7B · 6 benchmarks averageaccuracy: 68.5 -> 68.8No clear effect
apply STTM to larger or different MLLMs (OneVision, Qwen2VL, 72B)STTM · OneVision/Qwen2VL/LLaVA-Video-72Brelative accuracy: 102.1 (OneVision 50%), 102.7 (Qwen2VL 50%), 101.3 (72B 50%)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