Research map / Image generation / Super-resolution
Diffusion Models for Image Restoration and Enhancement: A Comprehensive Survey
Xin Li, Yulin Ren, Xin Jin, Cuiling Lan, Xingrui Wang, Wenjun Zeng, Xinchao Wang, Zhibo Chen
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 a residual-generating diffusion model instead of a plain diffusion SR model | ResShift vs SR3 · DIV2K 4x SR | PSNR: 20.37 -> 29.38 | Improved |
| GAN-based SR instead of diffusion SR for reference perceptual metrics | SeD vs ResShift · DIV2K 4x SR | FID: 15.41 -> 12.51 | Improved |
| deterministic transformer/Mamba SR instead of diffusion SR for fidelity | MambaIRv2 vs SUPIR · DIV2K 4x SR | PSNR: 27.19 -> 31.24 | Improved |
| diffusion SR built on pre-trained Stable Diffusion for real-world degradation | PASD vs RealESRGAN · RealSR | MUSIQ: 60.37 -> 68.69 | Improved |
| diffusion SR priors for real-world degradation hurts pixel fidelity | SUPIR vs CDC · RealSR | PSNR: 27.81 -> 23.47 | Hurt |
| fewer sampling steps via ResShift (15 steps) versus SR3 (100) | ResShift vs SR3 · DIV2K 4x SR | NFE: 100 -> 15 | Improved |
| zero-shot decomposition-based DDRM instead of DPS for pixel consistency | DDRM vs DPS · ImageNet 1K 4x SR | PSNR: 25.88 -> 27.38 | Improved |
| DPS-style sampling gives better perceptual LPIPS but is slow | DPS vs DDRM · ImageNet 1K 4x SR | LPIPS: 0.22 -> 0.15 | Improved |
| DDRM with 20 sampling steps instead of 1000 for DPS | DDRM vs DPS · ImageNet 1K 4x SR | time per image: 141.2 -> 10.1 | Improved |
| unsupervised training (LightenDiffusion) instead of paired training (DiffLL) | LightenDiffusion vs DiffLL · LOL | PSNR: 26.34 -> 20.45 | Hurt |
| Retinex decomposition conditioning in diffusion | Diff-Retinex vs GDP · LOL | PSNR: 15.90 -> 21.98 | Improved |
| unsupervised illumination-reflectance decoupling gives better generalization on unpaired real data | LightenDiffusion vs Diff-Retinex · NPE | NIQE: 4.996 -> 3.618 | Improved |
Closest papers by meaning
- Reconciling Stochastic and Deterministic Strategies for Zero-shot Image Restoration using Diffusion Model in Dual (CVPR 2025 · similarity 0.92)
- A Modular Conditional Diffusion Framework for Image Reconstruction (NeurIPS 2024 · similarity 0.92)
- Blind Image Restoration via Fast Diffusion Inversion (NeurIPS 2024 · similarity 0.91)
- Improving Diffusion-Based Image Restoration with Error Contraction and Error Correction (AAAI 2024 · similarity 0.91)
- Residual Denoising Diffusion Models (CVPR 2024 · similarity 0.91)
- IRBridge: Solving Image Restoration Bridge with Pre-trained Generative Diffusion Models (ICML 2025 · 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.
