Research map / Image generation / Super-resolution
One-Step Flow for Image Super-Resolution with Tunable Fidelity-Realism Trade-offs
Yuanzhi Zhu, Ruiqing Wang, Shilin Lu, Hanshu Yan, Junnan Li, Kai Zhang
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 Gaussian noise perturbation sigma_p=0.1 to LR initial state (ablation) | conditional flow · FFHQ 256 4x SR | FID: 110.29 -> 30.54 | Improved |
| increase sigma_p from 0.1 to 1 (ablation) | conditional flow · FFHQ 256 4x SR | NFEs / FID: NFE 20 -> 44; FID 30.54 -> 30.75 | Hurt |
| condition on LR image (ablation) | conditional flow · FFHQ 256 4x SR | FID: 42.47 -> 30.54 | Improved |
| use l1 loss instead of l2 (ablation) | conditional flow · FFHQ 256 4x SR | FID: 31.86 -> 30.54 | Improved |
| use proposed distillation loss instead of BOOT (ablation) | one-step student · FFHQ 256 4x SR | LPIPS: 0.483 -> 0.065 | Improved |
| use proposed loss instead of PINN loss (ablation) | one-step student · FFHQ 256 4x SR | LPIPS: 0.250 -> 0.065 | Improved |
| add alignment and boundary losses plus midpoint solver (ablation) | one-step student · FFHQ 256 4x SR | LPIPS: 0.065 -> 0.056 | Improved |
| reduce dt from 0.05 to 0.001 (ablation) | one-step student · FFHQ 256 4x SR | LPIPS: 0.065 -> 0.160 | Hurt |
| lower t from 1 to 0 at inference | one-step student · FFHQ 256 4x SR | PSNR vs LPIPS: PSNR 28.98 -> 31.25; LPIPS 0.055 -> 0.150 | Mixed |
| distill with one step instead of 20 NFE teacher | OFTSR · FFHQ 256 4x SR | LPIPS: 0.053 -> 0.055 | No clear effect |
| distill ResShift teacher with OFTSR loss instead of SinSR | one-step student · ImageNet 256 4x SR | FID: 94.90 -> 60.64 | Improved |
Closest papers by meaning
- Boosting Flow-based Generative Super-Resolution Models via Learned Prior (CVPR 2024 · similarity 0.91)
- SinSR: Diffusion-Based Image Super-Resolution in a Single Step (CVPR 2024 · similarity 0.91)
- One Diffusion Step to Real-World Super-Resolution via Flow Trajectory Distillation (ICML 2025 · similarity 0.91)
- TSD-SR: One-Step Diffusion with Target Score Distillation for Real-World Image Super-Resolution (CVPR 2025 · similarity 0.90)
- Noise-Started One-Step Image Super-Resolution via LR-Conditioned SplitMeanFlow and GAN Refinement (NeurIPS 2026 · similarity 0.89)
- Efficient Conditional Diffusion Model with Probability Flow Sampling for Image Super-resolution (AAAI 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.
