AI Paper Atlas

Research map / Video / Video generation

EVATok: Adaptive Length Video Tokenization for Efficient Visual Autoregressive Generation

Tianwei Xiong, Jun Hao Liew, Zilong Huang, Zhijie Lin, Jiashi Feng, Xihui Liu

CVPR 2026 · Video › Video generation · approach: Autoregressive long video generation · 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
assign tokens with the router instead of uniform 1024 tokens (no discriminator)final tokenizer · WebVid valrFVD: 63 -> 33 with 1024 -> 721 tokensImproved
assign tokens with the router instead of uniform 1024 tokens (+ VideoMAE discriminator)final tokenizer · WebVid valrFVD: 13 -> 9.2 with 1024 -> 721 tokensImproved
assign tokens with the router instead of uniform on UCFfinal tokenizer + GPT-B · UCF-101gFVD: 98 -> 96 with 1024 -> 740 generated tokensImproved
assign tokens with the router instead of uniform on UCFfinal tokenizer · UCF-101LPIPS: 0.1303 -> 0.1212Improved
train a final tokenizer from scratch instead of reusing the proxy tokenizer (uniform) (ablation)uniform tokenizer · WebVid valrFVD: 73 -> 63Improved
train a final tokenizer from scratch instead of reusing the proxy tokenizer (router) (ablation)router tokenizer · WebVid valrFVD: 50 -> 33Improved
add VideoMAE discriminator (ablation)final recipe · UCF-101rFVD: 65 -> 13Improved
add VideoMAE discriminator (ablation)final recipe · UCF-101gFVD: 155 -> 98Improved
add V-JEPA2 representation alignment (ablation)final recipe · UCF-101gFVD: 144 -> 98Improved
add both VideoMAE discriminator and V-JEPA2 alignment (ablation)final recipe · UCF-101rFVD: 80 -> 13Improved
add VideoMAE discriminatoruniform final tokenizer · WebVid valPSNR: 27.77 -> 26.68Hurt
use max-proxy-reward assignment or router instead of uniform assignmentproxy tokenizer · WebVid and UCFrFVD: 56% token savings on WebVid and 42% on UCFImproved

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