Research map / Image generation / Super-resolution
DNF-SR: Dual-Input and Negative-Aware Feature Fine-Tuning for Real-World Image Super-Resolution
Shuhao Han, Wenjie Liao, Hayden Vance, Hang Dong, Rui Zhang, Chun-Le Guo, Chongyi Li
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 dual input (z_mix and z_LR) instead of z_LR only (ablation) | DNF-SR (Flux) · RealSR | LPIPS: 0.3040 -> 0.2995 | Improved |
| add dual input instead of noisy z_mix only (ablation) | DNF-SR (Flux) · RealSR | LPIPS: 0.3349 -> 0.2995 | Improved |
| replace text-to-image Flux with image-editing Flux-Kontext (ablation) | DNF-SR dual input · RealSR | MANIQA: 0.6718 -> 0.6903 | Mixed |
| use mid-timestep t=0.5 instead of pure noise t=1 (ablation) | DNF-SR · RealSR | LPIPS: 0.30250 -> 0.2925 | Improved |
| use t=0.5 instead of t=0.25 (ablation) | DNF-SR · RealSR | MANIQA: 0.6870 -> 0.6903 | Improved |
| use t=0.5 instead of t=0.75 (ablation) | DNF-SR · RealSR | MANIQA: 0.6790 -> 0.6903 | Improved |
| add NF2T post-training (ablation) | DNF-SR(sft) · RealSR | QALIGN: 3.9162 -> 4.0718 | Improved |
| add NF2T post-training (ablation) | DNF-SR(sft) · RealSR | MUSIQ: 70.6717 -> 72.0396 | Improved |
| add NF2T post-training (ablation) | DNF-SR(sft) · RealSR | LPIPS: 0.2925 -> 0.3239 | Hurt |
| optimize in image and feature space (NF2T) instead of plain NFT (ablation) | DNF-SR · RealSR | MANIQA: 0.6413 -> 0.693 | Improved |
| add full-reference metrics to the NF2T reward (ablation) | DNF-SR · RealSR | MUSIQ: 71.0911 -> 72.0396 | Improved |
| DNF-SR vs OMGSR on RealLQ250 | OMGSR · RealLQ250 | CLIPIQA: 0.7435 -> 0.7997 | Improved |
Closest papers by meaning
- Unleashing the Power of One-Step Diffusion based Image Super-Resolution via a Large-Scale Diffusion Discriminator (NeurIPS 2025 · similarity 0.91)
- Noise-Started One-Step Image Super-Resolution via LR-Conditioned SplitMeanFlow and GAN Refinement (NeurIPS 2026 · similarity 0.91)
- One-Step Effective Diffusion Network for Real-World Image Super-Resolution (NeurIPS 2024 · similarity 0.90)
- SinSR: Diffusion-Based Image Super-Resolution in a Single Step (CVPR 2024 · similarity 0.90)
- ResDiff: Combining CNN and Diffusion Model for Image Super-resolution (AAAI 2024 · similarity 0.90)
- Taming Diffusion Prior for Image Super-Resolution with Domain Shift SDEs (NeurIPS 2024 · similarity 0.90)
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.
