Research map / Image generation / Autoregressive and latent generation
SoftCFG: Uncertainty-guided Stable Guidance for Visual Autoregressive Model
Dongli Xu, Aleksei Tiulpin, Matthew B. Blaschko
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 |
|---|---|---|---|
| replace vanilla CFG with SoftCFG (ablation) | AliTok-XL · ImageNet 256x256 | FID: 1.37 -> 1.32 | Improved |
| replace vanilla CFG with SoftCFG (ablation) | AliTok-XL · ImageNet 256x256 | IS: 321.4 -> 288.1 | Hurt |
| add Step Normalization to SoftCFG (ablation) | SoftCFG on AliTok-XL · ImageNet 256x256 | IS: 288.1 -> 302.0 | Improved |
| add Step Normalization to SoftCFG (ablation) | SoftCFG on AliTok-XL · ImageNet 256x256 | sFID: 7.62 -> 7.16 | Improved |
| add Step Normalization to SoftCFG (ablation) | SoftCFG on AliTok-XL · ImageNet 256x256 | FID: 1.32 -> 1.32 | No clear effect |
| jointly tune guidance scale and cosine exponent for SoftCFG (ablation) | SoftCFG + StepNorm on AliTok-XL · ImageNet 256x256 | FID: 1.32 -> 1.27 | Improved |
| use SoftCFG instead of CFG on AliTok-L | AliTok-L · ImageNet 256x256 | FID: 1.42 -> 1.39 | Improved |
| use SoftCFG instead of CFG on AliTok-XL | AliTok-XL · ImageNet 256x256 | IS: 321.4 -> 302.4 | Hurt |
| use SoftCFG instead of CFG across guidance scales (ablation) | AliTok-XL · ImageNet 256x256 | FID vs guidance scale | Improved |
Closest papers by meaning
- Toward Guidance-Free AR Visual Generation via Condition Contrastive Alignment (ICLR 2025 · similarity 0.89)
- Guiding Visual Autoregressive Models through Spectrum Weakening (NeurIPS 2026 · similarity 0.88)
- OmniGen-AR: AutoRegressive Any-to-Image Generation (NeurIPS 2025 · similarity 0.88)
- NFIG: Multi-Scale Autoregressive Image Generation via Frequency Ordering (NeurIPS 2025 · similarity 0.88)
- Vision Foundation Models as Effective Visual Tokenizers for Autoregressive Generation (NeurIPS 2025 · similarity 0.87)
- Visual Autoregressive Modeling: Scalable Image Generation via Next-Scale Prediction (NeurIPS 2024 · similarity 0.87)
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.
