AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

AViTS:Adaptive Spatiotemporal Token Selection for Efficient Dynamic-Resolution Generation

Haoran Qin, zhengan yan, Shikang Zheng, Xiaobing Tu, Jiacheng Liu, Yuqi Lin, Chang Zou, Jinshan Liu, Peiliang Cai, Xiantao Zhang, Jinkui Ren, Linfeng Zhang

ECCV 2026 · Image generation › Autoregressive and latent generation · approach: Generative latent space tooling

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 step-variance temporal importance to attention-only selection (ablation)AViTS · DrawBench NFE 30ImageReward: 0.9875 -> 0.9959Improved
add attention spatial importance to step-variance-only selection (ablation)AViTS · DrawBench NFE 30ImageReward: 0.9846 -> 0.9959Improved
use importance-based selection instead of random token selection (ablation)AViTS · DrawBench NFE 30ImageReward: 0.9257 -> 0.9959Improved
use importance-based selection instead of even token selection (ablation)AViTS · DrawBench NFE 30CLIP Score: 32.012 -> 32.361Improved
use attention+variance instead of edge detection (ablation)AViTS · DrawBench NFE 30ImageReward: 0.9512 -> 0.9959Improved
AViTS replaces RALU at NFE 30dynamic-resolution sampling · GEdit-EN Qwen-Image-EditOverall Score: 7.52 -> 7.62Improved
AViTS replaces TaylorSeer at similar speedupFLUX.1-Kontext-dev · GEdit-ENOverall Score: 6.17 -> 6.70Improved
AViTS replaces FORA at similar speedupQwen-Image-Edit · GEdit-ENOverall Score: 7.28 -> 7.55Improved
apply AViTS at 1440x1440 instead of direct samplingFLUX.1-dev · DrawBench 1440x1440ImageReward: 0.7385 -> 0.9008Improved
apply AViTS at 2048x2048 instead of direct samplingFLUX.1-dev · DrawBench 2048x2048ImageReward: 0.8371 -> 0.8905Improved
increase upsampling ratio rho (ablation)AViTS · DrawBenchImageRewardMixed
use either attention-only (alpha=1) or variance-only (alpha=0) rather than intermediate alpha (ablation)AViTS · FLUX.1-dev, Qwen-Image-Edit, KontextImageReward / OSHurt

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