AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

STAT: Soft Tail-dropping for Adaptive Visual Tokenization

Zeyuan Chen, Kai Zhang, Zhuowen Tu, Yuanjun Xiong

ECCV 2026 · Image generation › Autoregressive and latent generation · approach: Next-scale and token autoregressive images

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 adaptive token allocation instead of a fixed token count (220) (ablation)STAT tokenizer · ImageNet-1k class-conditional generation (LlamaGen-XL)gFID: 2.73 -> 2.36Improved
use adaptive token allocation instead of a fixed token count (220) (ablation)STAT tokenizer · ImageNet-1k val 256x256 reconstructionrFID: 1.15 -> 1.15No clear effect
use adaptive token allocation instead of a fixed token count (220) (ablation)STAT tokenizer · ImageNet-1k val 256x256 reconstructionPSNR: 20.35 -> 20.22Hurt
use per-token probabilistic (Bernoulli) dropping instead of hard tail-drop (ablation)STAT tokenizer · ImageNet-1k class-conditional generation (LlamaGen-XL)gFID: 2.67 -> 2.36Improved
use per-token probabilistic (Bernoulli) dropping instead of hard tail-drop (ablation)STAT tokenizer · ImageNet-1k val 256x256 reconstructionPSNR: 20.30 -> 20.22Hurt
sample the EoS threshold during AR training instead of a fixed threshold (ablation)LlamaGen-STAT-XL · ImageNet-1k class-conditional generation (LlamaGen-XL)gFID: 2.49 -> 2.36Improved
add content-adaptive loss (ablation)STAT tokenizer · ImageNet-1k reconstructiontoken allocationImproved
add decreasing importance loss (ablation)STAT tokenizer · ImageNet-1k reconstructiontail-dropping patternImproved
lower the inference threshold from 0.5 to 0.01STAT · ImageNet-1k val 256x256 reconstructionrFID: 1.15 -> 0.99Improved
add 10 extra tokens at inference on top of threshold 0.01STAT · ImageNet-1k val 256x256 reconstructionrFID: 0.99 -> 0.88Improved
use expected token count instead of threshold 0.5 at inference (ablation)STAT · ImageNet-1k val 256x256 reconstructionrFID: 1.15 -> 1.20No clear effect
replace the LlamaGen VQ tokenizer with STAT in text-conditional ARvanilla AR model · GenEvalOverall: 0.32 -> 0.45Improved

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