Research map / Image generation / Autoregressive and latent generation
SJD-SV: Speculative Jacobi Decoding with Semantics Verification for Autoregressive Image Generation
Baoquan Zhang, Bingqi Shan, Shihao Fang, Kenghong Lin, Xutao Li, Yunming Ye
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 changed | Where | Measured | Effect |
|---|---|---|---|
| add SJD-SV to SJD | SJD on Lumina-mGPT · Parti-prompt | NFE: 1035.3 -> 826.32 | Improved |
| add SJD-SV to GSD | GSD on Lumina-mGPT · Parti-prompt | NFE: 898.97 -> 721.47 | Improved |
| add SJD-SV to LANTERN | LANTERN on Lumina-mGPT · MS-COCO 2017 | latency: 33.80s -> 23.11s | Improved |
| add SJD-SV to SJD | SJD on Lumina-mGPT · Parti-prompt | CLIP-Score: 32.09 -> 32.13 | No clear effect |
| add progressive fallback (ablation) | SJD-SV · Parti-prompt | NFE: 972.92 -> 826.32 | Improved |
| use adaptive suffix instead of all tokens for joint probability (ablation) | SJD-SV · Parti-prompt | NFE: 877.29 -> 826.32 | Improved |
| verify only the last token instead of adaptive suffix (ablation) | SJD-SV · Parti-prompt | CLIP-Score: 32.13 -> 31.04 | Hurt |
| lower alpha from 0.8 to 0.5 (ablation) | SJD-SV · Parti-prompt | NFE / CLIP-Score: NFE 826.32 -> 670.80; CLIP 32.13 -> 31.88 | Mixed |
| use alpha 0.8 rather than 1.0 (full product) (ablation) | SJD-SV · Parti-prompt | NFE: 877.29 -> 826.32 | Improved |
| add SJD-SV to SJD (ablation) | SJD on Lumina-mGPT · Parti-prompt, ambiguous positions | acceptance rate: 56.49% -> 62.03% | Improved |
| add SJD-SV to SJD | SJD on Emu3 · Parti-prompt | latency acceleration: 2.26x -> 4.33x | Improved |
| add SJD-SV to SJD | SJD on LlamaGen · Parti-prompt | NFE: 569.0 -> 555.4 | Improved |
Closest papers by meaning
- Speculative Jacobi-Denoising Decoding for Accelerating Autoregressive Text-to-image Generation (NeurIPS 2025 · similarity 0.92)
- Accelerating Auto-regressive Text-to-Image Generation with Training-free Speculative Jacobi Decoding (ICLR 2025 · similarity 0.90)
- LANTERN: Accelerating Visual Autoregressive Models with Relaxed Speculative Decoding (ICLR 2025 · similarity 0.86)
- Towards Better & Faster Autoregressive Image Generation: From the Perspective of Entropy (NeurIPS 2025 · similarity 0.86)
- Vision Foundation Models as Effective Visual Tokenizers for Autoregressive Generation (NeurIPS 2025 · similarity 0.85)
- Epsilon-VAE: Denoising as Visual Decoding (ICML 2025 · similarity 0.85)
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.
