Research map / Image generation / Super-resolution
You Only Need One Step: Fast Super-Resolution with Stable Diffusion via Scale Distillation
Mehdi Noroozi, Isma Hadji, Brais Martinez, Adrian Bulat, Georgios Tzimiropoulos
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 scale distillation (x2 to x4) to the SR diffusion model (ablation) | SD-based SR, 1 step, original decoder · DIV2K bicubic x4 | FID: 27.93 -> 21.63 | Improved |
| add scale distillation (x2 to x4 to x8) (ablation) | SD-based SR, 1 step, original decoder · DIV2K bicubic x8 | FID: 102.92 -> 56.84 | Improved |
| fine-tune the decoder on top of the frozen one-step U-Net (ablation) | YONOS-SR · DIV2K bicubic x4 | FID: 21.63 -> 12.25 | Improved |
| fine-tune the decoder on top of the frozen one-step U-Net (ablation) | YONOS-SR · DIV2K bicubic x8 | FID: 56.84 -> 21.48 | Improved |
| add scale distillation before decoder fine-tuning (ablation) | SD-SR with fine-tuned decoder · DIV2K bicubic x8 | FID: 41.54 -> 21.48 | Improved |
| use scale distillation with few inference steps (ablation) | SD-SR x4 · DIV2K bicubic x4 | FID: at least 6 points better at 1 step | Improved |
| scale distillation benefit shrinks as inference steps increase (ablation) | SD-SR x4 · DIV2K bicubic x4 | FID | Mixed |
| use 1-step YONOS-SR instead of 200-step StableSR | StableSR · DIV2K Valid (RealESRGAN degradations) | MUSIQ: 65.92 -> 68.35 | Improved |
| 1-step YONOS-SR vs P2L on FFHQ x8 | YONOS-SR · FFHQ x8 | PSNR: 28.78 -> 26.08 (vs LDPS) | Hurt |
Closest papers by meaning
- SinSR: Diffusion-Based Image Super-Resolution in a Single Step (CVPR 2024 · similarity 0.93)
- TSD-SR: One-Step Diffusion with Target Score Distillation for Real-World Image Super-Resolution (CVPR 2025 · similarity 0.92)
- Unleashing the Power of One-Step Diffusion based Image Super-Resolution via a Large-Scale Diffusion Discriminator (NeurIPS 2025 · similarity 0.92)
- Arbitrary-steps Image Super-resolution via Diffusion Inversion (CVPR 2025 · similarity 0.91)
- Taming Diffusion Prior for Image Super-Resolution with Domain Shift SDEs (NeurIPS 2024 · similarity 0.91)
- Solving Diffusion ODEs with Optimal Boundary Conditions for Better Image Super-Resolution (ICLR 2024 · similarity 0.91)
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.
