AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

SpectralAR: Spectral Autoregressive Visual Generation

Yuanhui Huang, Weiliang Chen, Wenzhao Zheng, Yueqi Duan, Jie Zhou, Jiwen Lu

ICCV 2025 · Image generation › Autoregressive and latent generation · approach: Next-scale and token autoregressive images · 12 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 spectral (sub-image) supervision to the 1D tokenizer (ablation)SpectralAR-d16 (TiTok baseline) · ImageNet-1K 256x256 gFIDgFID: 6.30 -> 5.64Improved
add spectral causal mask in encoder and decoder (ablation)SpectralAR-d16 · ImageNet-1K 256x256 gFIDgFID: 5.64 -> 3.49Improved
add non-uniform token-frequency mapping (ablation)SpectralAR-d16 · ImageNet-1K 256x256 gFIDgFID: 3.49 -> 3.02Improved
causal mask without non-uniform mapping vs spectral supervision only, effect on IS (ablation)SpectralAR-d16 · ImageNet-1K 256x256 ISIS: 255.1 -> 222.6Mixed
use spectral token sequence instead of raster-scan orderCIFAR-100 token sequences · CIFAR-100 val causality proxyaverage R^2: 0.454 -> 0.962Improved
spectral sequence versus scale-wise sequenceCIFAR-100 token sequences · CIFAR-100 val causality proxyaverage R^2: 0.928 -> 0.962Improved
truncate last 5 generated tokens (ablation)SpectralAR-d16 · ImageNet-1K 256x256FID: 3.02 -> 3.34Hurt
truncate last 15 generated tokens (ablation)SpectralAR-d16 · ImageNet-1K 256x256FID: 3.02 -> 27.69Hurt
refine strided sub-images with SpectralAR for super-resolutionupsampled 512x512 images · ImageNet super-resolutionFID: 3.09 -> 2.93Improved
use patch-based sub-images instead of strided for super-resolution (ablation)SpectralAR super-resolution · ImageNet super-resolutionFID: 2.93 -> 14.76Hurt
replace TiTok 1D AR tokenizer with SpectralAR tokenization310M AR generator, 64 tokens · ImageNet-1K 256x256gFID: 6.30 -> 3.02Improved
nested spectral tokenization instead of plain 1D tokenizationtokenizer reconstruction · ImageNet-1K valrFID: 1.70 -> 4.03Hurt

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