Research map / Image generation / Super-resolution
One-Step Diffusion Transformer for Controllable Real-World Image Super-Resolution
Yushun Fang, Yuxiang Chen, Shibo Yin, Qiang Hu, Jiangchao Yao, Ya Zhang, Xiaoyun Zhang, Yanfeng 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 |
|---|---|---|---|
| add a separate Control Noise visual stream (NVS) instead of a single visual stream (ablation) | ODTSR · RealSR (f=1.0, w/o prompt) | FID: 118.08 -> 101.49 | Improved |
| add NVS instead of single visual stream (ablation) | ODTSR · RealSR (f=1.0, w/o prompt) | LPIPS: 0.2655 -> 0.2398 | Improved |
| add NVS at strong controllability setting (ablation) | ODTSR · RealSR (f=0.2, w/ prompt) | LPIPS: 0.2915 -> 0.3212 | Mixed |
| add NVS at strong controllability setting (ablation) | ODTSR · RealSR (f=0.2, w/ prompt) | CLIP-T: 34.35 -> 34.56 | Improved |
| use fidelity-aware adversarial weight instead of fixed weight 0.02 (ablation) | ODTSR · RealSR (f=0.2, with prompt) | CLIP-T: 33.09 -> 34.56 | Improved |
| use fidelity-aware adversarial weight instead of fixed weight 0.1 (ablation) | ODTSR · RealSR (f=0.2, with prompt) | CLIP-T: 32.71 -> 34.56 | Improved |
| use fidelity-aware adversarial weight instead of fixed weight 0.02 (ablation) | ODTSR · RealSR (f=1.0, with prompt) | MANIQA: 0.646 -> 0.668 | Improved |
| lower fidelity weight f from 1.0 to 0.2 | ODTSR · RealSR (with prompt) | CLIP-T: 34.01 -> 34.56 | Improved |
| lower fidelity weight f from 1.0 to 0.2 | ODTSR · RealSR (with prompt) | PSNR: 25.09 -> 21.70 | Hurt |
| add the text prompt to ODTSR on scene text | ODTSR (f=1.0) · RealCE-val | NED: 0.7609 -> 0.8475 | Improved |
| add the text prompt to ODTSR on generic RealSR | ODTSR (f=1.0) · RealSR | FID: 101.49 -> 91.12 | Improved |
| use one-step ODTSR instead of multi-step DiT4SR at low CFG | DiT4SR · RealSR (with prompt) | MUSIQ: 61.94 -> 68.22 | Improved |
Closest papers by meaning
- Adversarial Diffusion Compression for Real-World Image Super-Resolution (CVPR 2025 · similarity 0.90)
- Effective Diffusion Transformer Architecture for Image Super-Resolution (AAAI 2025 · similarity 0.90)
- Noise-Started One-Step Image Super-Resolution via LR-Conditioned SplitMeanFlow and GAN Refinement (NeurIPS 2026 · similarity 0.90)
- When Latents Forget Pixels: Restoring Fidelity in Diffusion Transformer Super-Resolution (NeurIPS 2026 · similarity 0.89)
- One Diffusion Step to Real-World Super-Resolution via Flow Trajectory Distillation (ICML 2025 · similarity 0.89)
- SinSR: Diffusion-Based Image Super-Resolution in a Single Step (CVPR 2024 · similarity 0.89)
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.
