AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

Spectral Image Tokenizer

Carlos Esteves, Mohammed Suhail, Ameesh Makadia

ICCV 2025 · Image generation › Autoregressive and latent generation · approach: Next-scale and token autoregressive images · 11 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
use DWT spectral tokens (SIT-5) instead of ViT-VQGAN patch tokensViT-VQGAN · ImageNet 256x256 reconstructionFID: 2.33 -> 0.97Improved
use SIT-6 instead of ViT-VQGAN at 512x512ViT-VQGAN · ImageNet 512x512 reconstructionFID: 6.915 -> 1.74Improved
add scale-causal attention to encoder and decoder (SIT-SC-5) (ablation)SIT-5 · ImageNet 256x256 reconstructionFID: 0.97 -> 1.33Hurt
add scale-causal decoder attention (ablation)SIT-4 · ImageNet 256x256 reconstructionFID: 1.20 -> 1.45Hurt
add scale-causal attention on encoder and decoder (ablation)SIT-4 · ImageNet 256x256 reconstructionFID: 1.20 -> 1.97Hurt
use LeGall5/3 wavelet instead of Haar (ablation)SIT-5 · ImageNet reconstructionFID: 0.97 -> 1.10Hurt
use CDF9/7 wavelet instead of Haar (ablation)SIT-5 · ImageNet reconstructionFID: 0.97 -> 1.14Hurt
reduce vocabulary from 8192 to 4096 (ablation)SIT-5 · ImageNet reconstructionFID: 0.97 -> 1.33Hurt
increase vocabulary from 8192 to 16384 (ablation)SIT-5 · ImageNet reconstructionFID: 0.97 -> 0.91Improved
use SIT-2 with sequence length 2048 instead of SIT-4 with 1024 (ablation)SIT · ImageNet reconstructionFID: 1.20 -> 0.77Improved
use SIT-SC-5 instead of ViT-VQGAN at 64x64 inputViT-VQGAN · ImageNet 64x64 reconstructionFID: 3.53 -> 1.39Improved
use AR-SIT-SCD-4 instead of Parti350M at 256x256Parti350M · MS-COCO text-to-imageFID: 12.35 -> 12.57No clear effect

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