Research map / Image generation / Fast diffusion sampling
$\sigma$: Sigmoid Modulation for Ultra High Resolution Diffusion
Bingxuan Zhao, Qing Zhou, Yu Wang, Chuang Yang, Qi Wang
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 |
|---|---|---|---|
| use tc=1/s center instead of constant tc=1 (ablation) | SigMa-YaRN · Aesthetic-4K (4x square) | FID: 199.50 -> 184.73 | Improved |
| use tc=1/s center instead of 1/s^2 (ablation) | SigMa-YaRN · Aesthetic-4K (4x square) | FID: 208.30 -> 184.73 | Improved |
| use tc=1/s center instead of 1/sqrt(s) (ablation) | SigMa-YaRN · Aesthetic-4K (4x square) | FID: 189.10 -> 184.73 | Improved |
| use gamma=sqrt(s) instead of constant gamma=1 (ablation) | SigMa-YaRN · Aesthetic-4K (4x square) | FID: 194.20 -> 184.73 | Improved |
| use gamma=sqrt(s) instead of gamma=1/s (ablation) | SigMa-YaRN · Aesthetic-4K (4x square) | HPSv2: 0.245 -> 0.287 | Improved |
| use gamma=sqrt(s) instead of gamma=s (ablation) | SigMa-YaRN · Aesthetic-4K (4x square) | FID: 191.40 -> 184.73 | Improved |
| replace DyPE power-law schedule with SigMa sigmoid schedule | YaRN · Aesthetic-4K (4x square) | FID: 188.53 -> 184.73 | Improved |
| replace DyPE power-law schedule with SigMa sigmoid schedule | NTK-aware · Aesthetic-4K (4x square) | FID: 184.99 -> 185.35 | No clear effect |
| replace DyPE power-law schedule with SigMa sigmoid schedule | YaRN · Panoramic 4096x1024 | HPSv2: 0.231 -> 0.254 | Improved |
| add time-dependent DyPE modulation to static YaRN | YaRN · Aesthetic-4K (4x square) | HPSv2: 0.261 -> 0.279 | Improved |
| replace DyPE with SigMa on a second backbone | YaRN · Aesthetic-4K (Qwen-Image) | HPSv2: 0.277 -> 0.281 | Improved |
| apply static NTK-aware interpolation to FLUX | FLUX.1-dev · DrawBench (4x square) | HPSv2: 0.139 -> 0.118 | Hurt |
Closest papers by meaning
- ScaleDiff: Higher-Resolution Image Synthesis via Efficient and Model-Agnostic Diffusion (NeurIPS 2025 · similarity 0.89)
- FAM Diffusion: Frequency and Attention Modulation for High-Resolution Image Generation with Stable Diffusion (CVPR 2025 · similarity 0.89)
- SEGA: Spectral-Energy Guided Attention for Resolution Extrapolation in Diffusion Transformers (NeurIPS 2026 · similarity 0.89)
- Simpler Diffusion: 1.5 FID on ImageNet512 with Pixel-space Diffusion (CVPR 2025 · similarity 0.89)
- EfficientDM: Efficient Quantization-Aware Fine-Tuning of Low-Bit Diffusion Models (ICLR 2024 · similarity 0.88)
- Faster Diffusion: Rethinking the Role of the Encoder for Diffusion Model Inference (NeurIPS 2024 · similarity 0.88)
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.
