AI Paper Atlas

Research map / Image generation / Autoregressive and latent generation

SJD++: Improved Speculative Jacobi Decoding for Training-Free Acceleration of Discrete Auto-Regressive Text-to-Image Generation

Yao Teng, Zhihuan Jiang, Han Shi, Xihui Liu, Xuefei Ning, Guohao DaiMember, Yu Wang, Zhenguo Li, Xihui Liu

TPAMI 2026 · Image generation › Autoregressive and latent generation · approach: Accelerating autoregressive image generation · 3 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 speculative Jacobi decoding with token reuse (SJD++) vs SJD (ablation)Lumina-mGPT-7B · MS-COCO2017 vallatency acceleration: 2.04x -> 3.12x; step 2.22x -> 6.44xImproved
add token reuse (SJD++) vs SJD (ablation)Emu3-Gen-8B · MS-COCO2017 vallatency acceleration: 1.81x -> 3.01x; step 2.01x -> 7.50xImproved
add token reuse (SJD++) vs SJD (ablation)Janus-Pro-1B · MS-COCO2017 vallatency acceleration: 1.81x -> 2.31xImproved
add token reuse (SJD++) vs SJD (ablation)SimpleAR-0.5B · MS-COCO2017 vallatency acceleration: 1.94x -> 2.78xImproved
add token reuse (SJD++) vs SJD (ablation)Lumina-mGPT-7B · MS-COCO2017 valFID: 30.90 -> 31.48No clear effect
use SJD++ instead of EAGLE or LANTERN or ZipARLumina-mGPT · MS-COCO2017 vallatency acceleration: 2.56x -> 3.12x (LANTERN vs SJD++)Improved
add Jacobi window size 96 instead of 16 (ablation)SJD++ Janus-Pro-1B · GenevalGeneval score: 0.7060 -> 0.7248Improved
use Jacobi window size 32 instead of 96 (ablation)SJD++ Janus-Pro-1B · GenevalGeneval score: 0.7248 -> 0.6838Hurt
disable refinement (replace unaccepted tokens with new initializations) (ablation)SJD · image generationforward passes: roughly one thousand -> more than two thousandHurt
use spatial-prior token initialization instead of random (ablation)SJD · simple-pattern promptCLIP-Score: 31.37 -> 31.52No clear effect
increase resolution from 512 to 1024 (ablation)SJD++ Lumina-mGPT-7B · Lumina-mGPT resolutionsstep compression: up to 2.43x at 1024x1024Improved
use Jacobi decoding instead of SJD under stochastic top-KJacobi Decoding · COCO2017 Lumina-mGPTlatency acceleration: 1.02x vs 2.05x for SJDNo 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