AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

Neighboring Autoregressive Modeling for Efficient Visual Generation

Yefei He, Yuanyu He, Shaoxuan He, Feng Chen, Hong Zhou, Kaipeng Zhang, Bohan Zhuang

ICCV 2025 · Image generation › Autoregressive and latent generation · approach: Next-scale and token autoregressive images · 30 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 dimension-oriented decoding heads (ablation)NAR-L · ImageNet 256x256FID: 66.31 -> 3.06Improved
mix logits of horizontal and vertical heads on overlapped tokens (ablation)NAR-L · ImageNet 256x256FID: 75.91 -> 3.06 (horizontal only), 25.73 -> 3.06 (vertical only)Improved
mix logits of horizontal and vertical heads on overlapped tokens (ablation)NAR-L · ImageNet 256x256IS: 22.36 -> 263.9 (horizontal only), 96.08 -> 263.9 (vertical only)Improved
replace next-token prediction with next-neighbor predictionLlamaGen-L · ImageNet 256x256FID and steps: 3.80 -> 3.06, 256 -> 31 stepsImproved
NAR-L (372M) vs LlamaGen-XXL (1.4B)LlamaGen-XXL · ImageNet 256x256throughput: 14.1 -> 195.4 img/sImproved
replace multi-scale next-scale prediction with NARVAR-d16 · ImageNet 256x256FID and throughput: 3.30 -> 3.27, 129.3 -> 248.5 img/sImproved
replace PAR-4X parallel decoding with NARPAR-XL-4X · ImageNet 256x256FID: 3.50 -> 2.70Improved
use NAR instead of LARP-L-Long autoregressionLARP-L-Long · UCF-101FVD and latency: reducing FVD by 5.8 and generation latency by 97.5%Improved
use NAR-XL instead of PAR-XL parallel decodingPAR-XL · UCF-101FVD: 23~32.3 FVD reductionImproved
increase PAR parallelism from 1x to 16xPAR-XL · UCF-101FVD: 94.1 -> 99.5 -> 103.4Hurt
scale NAR-L to deeper NAR-LPNAR-L · UCF-101FVD: 96.2 -> 71.1Improved
swap LlamaGen-XL for NAR-XL in text-to-imageLlamaGen-XL · GenEvaloverall score: 0.32 -> 0.43Improved

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