Research map / Image generation / Super-resolution
GenDR: Lighten Generative Detail Restoration
Yan Wang, Shijie Zhao, Kexin Zhang, Junlin Li, Li 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 |
|---|---|---|---|
| use CiD instead of VSD for step distillation (ablation) | GenDR · RealSR | MUSIQ: 65.01 -> 66.85 | Improved |
| use CiD instead of SiD for step distillation (ablation) | GenDR · RealSet80 | MUSIQ: 69.33 -> 70.61 | Improved |
| add adversarial loss to CiD (ablation) | GenDR · RealSR | ClipIQA / LPIPS: ClipIQA 0.6532 -> 0.7020; LPIPS 0.2527 -> 0.2899 | Mixed |
| add representation alignment (REPA) loss to CiD+adv (ablation) | GenDR · RealSR | MUSIQ: 68.30 -> 68.36 | No clear effect |
| use CiDA instead of VSD (ablation) | OSEDiff and GenDR · Q-Align | Q-Align: +0.08 | Improved |
| use 16-channel VAE (SD2.1-VAE16) instead of 4-channel VAE for super-resolution (ablation) | GenDR base model · RealSR / RealSet80 | IQA scores | Improved |
| use 16-channel VAE instead of 4-channel for text-to-image generation (ablation) | SD2.1 · COCO 2014 20k prompts | FID: 13.45 -> 27.89 | Hurt |
| replace text encoder and scheduler with fixed prompt embeddings (ablation) | GenDR pipeline · RealLR200 | ClipIQA / MUSIQ | No clear effect |
| use N3 negative prompt versus other negative prompts (ablation) | GenDR fixed prompts · RealLR | ClipIQA / MUSIQ: average 0.7231/71.58 vs Null 0.7193/71.40 | Improved |
| use LoRA and shared base model for score networks with CiDA (ablation) | CiDA training · training cost | time per iteration: memory comparable to VSD; training steps slower by 34% (CiD) and 42% (CiDA) | Mixed |
| GenDR one-step versus DreamClear multi-step | GenDR · inference speed | speedup: about 89.5x acceleration with half the parameters | Improved |
Later papers that compared with this method
1 later accepted paper reports GenDR-1 as a baseline in their results tables; 0 of them report a better number on most of the results they share with it.
- Eliminating VAE for Fast and High-Resolution Generative Detail Restoration (ICLR 2026 · better on 2 of 4 shared results)
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- SeeSR-50 — SeeSR: Towards Semantics-Aware Real-World Image Super-Resolution (CVPR 2024) · better on 1 of 1 shared result
- SinSR-1 — SinSR: Diffusion-Based Image Super-Resolution in a Single Step (CVPR 2024) · better on 0 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- RAP-SR: RestorAtion Prior Enhancement in Diffusion Models for Realistic Image Super-Resolution (AAAI 2025 · similarity 0.89)
- When Latents Forget Pixels: Restoring Fidelity in Diffusion Transformer Super-Resolution (NeurIPS 2026 · similarity 0.89)
- TSD-SR: One-Step Diffusion with Target Score Distillation for Real-World Image Super-Resolution (CVPR 2025 · similarity 0.89)
- Reconstruction vs. Generation: Taming Optimization Dilemma in Latent Diffusion Models (CVPR 2025 · similarity 0.89)
- Taming Diffusion Prior for Image Super-Resolution with Domain Shift SDEs (NeurIPS 2024 · similarity 0.89)
- Noise-Started One-Step Image Super-Resolution via LR-Conditioned SplitMeanFlow and GAN Refinement (NeurIPS 2026 · 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.
