AI Paper Atlas

Research map / Video / Video-language understanding

EarlyTom: Early Token Compression Completes Fast Video Understanding

Hesong Wang, Xin Jin, Lu Lu, Chenhaowen Li, Jian Chen, Qiang Liu, Huan Wang

CVPR 2026 · Video › Video-language understanding · approach: Video LLMs with fine-grained understanding · 4 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 decoupled spatial token selection to frame merging (ablation)EarlyTom, LLaVA-OV-7B · 3 video benchmarks avg, 20% retainedavg score: 58.4 -> 58.8Improved
add inner-encoder frame merge to spatial selection (ablation)EarlyTom, LLaVA-OV-7B · 3 video benchmarks avg, 20% retainedavg score: 58.4 -> 58.8Improved
local window sampling instead of random sampling (ablation)EarlyTom, LLaVA-OV-7B · 3 video benchmarks avg, 20% retainedavg score: 57.8 -> 58.8Improved
local window sampling instead of global top-K (ablation)EarlyTom, LLaVA-OV-7B · 3 video benchmarks avg, 20% retainedavg score: 58.4 -> 58.8Improved
local window sampling instead of global top-K (ablation)EarlyTom, LLaVA-OV-7B · throughput (tokens/s)throughput: 31.5 -> 33.4Improved
local window sampling instead of random sampling (ablation)EarlyTom, LLaVA-OV-7B · throughput (tokens/s)throughput: 35.3 -> 33.4Hurt
start frame merging at layer 6 instead of layer 4 (ablation)EarlyTom, LLaVA-OV-7B · avg score, 20% retainedavg score: 58.6 -> 58.9Improved
start frame merging at layer 10 instead of layer 6 (ablation)EarlyTom, LLaVA-OV-7B · TTFT (ms), 20% retainedTTFT: 387.1 -> 436.9Hurt
add decoupled spatial token selection (ablation)EarlyTom, Qwen2.5-VL-7B · avg score, 15% retainedavg score: 61.4 -> 62.2Improved
add weighted frame merging (ablation)EarlyTom, Qwen2.5-VL-7B · avg score, 15% retainedavg score: 61.3 -> 62.2Improved
compress inside vision encoder vs HoliTom late compressionEarlyTom, LLaVA-OV-7B · TTFT (ms), 10% retainedTTFT: 556.6 -> 336.2Improved
EarlyTom vs HoliTom accuracyEarlyTom, LLaVA-OV-7B · 4-benchmark avg, 10% retainedavg score: 57.9 -> 56.2Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 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