AI Paper Atlas

Research map / Image generation / Image restoration

Exploiting Diffusion Prior for Task-driven Image Restoration

Jaeha Kim, Junghun Oh, Kyoung Mu Lee

ICCV 2025 · Image generation › Image restoration · approach: Diffusion-based universal restoration · 7 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
add HLF loss (features of both task networks) (ablation)EDTR-1 step · Classification Mixture-Baccuracy: 67.8 -> 68.8Improved
add feature-matching (FM) loss (ablation)EDTR-1 step · Classification Mixture-Baccuracy: 67.6 -> 68.8Hurt
add partial diffusion (ablation)EDTR-1 step · Classification Mixture-Baccuracy: 63.5 -> 66.6Improved
add pre-restoration (SwinIR) before diffusion (ablation)EDTR-1 step · Classification Mixture-Baccuracy: 66.6 -> 68.8Improved
add HLF loss to SD at 50 steps (ablation)EDTR · Classification Mixture-Baccuracy: 64.3 -> 62.9Hurt
use 1 denoising step instead of 50 (ablation)EDTR with HLF loss · Classification Mixture-Baccuracy: 62.9 -> 63.5Improved
use Stable Diffusion prior (ablation)EDTR-1 step · Classification Mixture-Baccuracy: 64.2 -> 68.8Improved
increase denoising steps from 4 to 50 (ablation)EDTR · Classification Mixture-Baccuracy: 68.4 -> 66.8Hurt
increase denoising steps from 1 to 4 (ablation)EDTR · Semantic segmentationmIoU: 60.4 -> 62.9Improved
add trainable VAE decoder (ablation)EDTR-1 step · Classification Mixture-Baccuracy: 67.7 -> 68.8Improved
add wavelet color correction (ablation)EDTR-1 step · Classification Mixture-Baccuracy: 68.4 -> 68.8Improved
use SwinIR instead of EDSR for pre-restoration (ablation)EDTR-1 step · Classification Mixture-Baccuracy: 66.8 -> 68.8Improved

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