AI Paper Atlas

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

IJCV 2025 · Image generation › Super-resolution · approach: Diffusion-based super-resolution · 232 citations (OpenAlex)

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 changedWhereMeasuredEffect
use a residual-generating diffusion model instead of a plain diffusion SR modelResShift vs SR3 · DIV2K 4x SRPSNR: 20.37 -> 29.38Improved
GAN-based SR instead of diffusion SR for reference perceptual metricsSeD vs ResShift · DIV2K 4x SRFID: 15.41 -> 12.51Improved
deterministic transformer/Mamba SR instead of diffusion SR for fidelityMambaIRv2 vs SUPIR · DIV2K 4x SRPSNR: 27.19 -> 31.24Improved
diffusion SR built on pre-trained Stable Diffusion for real-world degradationPASD vs RealESRGAN · RealSRMUSIQ: 60.37 -> 68.69Improved
diffusion SR priors for real-world degradation hurts pixel fidelitySUPIR vs CDC · RealSRPSNR: 27.81 -> 23.47Hurt
fewer sampling steps via ResShift (15 steps) versus SR3 (100)ResShift vs SR3 · DIV2K 4x SRNFE: 100 -> 15Improved
zero-shot decomposition-based DDRM instead of DPS for pixel consistencyDDRM vs DPS · ImageNet 1K 4x SRPSNR: 25.88 -> 27.38Improved
DPS-style sampling gives better perceptual LPIPS but is slowDPS vs DDRM · ImageNet 1K 4x SRLPIPS: 0.22 -> 0.15Improved
DDRM with 20 sampling steps instead of 1000 for DPSDDRM vs DPS · ImageNet 1K 4x SRtime per image: 141.2 -> 10.1Improved
unsupervised training (LightenDiffusion) instead of paired training (DiffLL)LightenDiffusion vs DiffLL · LOLPSNR: 26.34 -> 20.45Hurt
Retinex decomposition conditioning in diffusionDiff-Retinex vs GDP · LOLPSNR: 15.90 -> 21.98Improved
unsupervised illumination-reflectance decoupling gives better generalization on unpaired real dataLightenDiffusion vs Diff-Retinex · NPENIQE: 4.996 -> 3.618Improved

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy