AI Paper Atlas

Research map / Image generation / Image restoration

Image Restoration via Diffusion Models with Dynamic Resolution

Yang Zheng, Wen Li, Zhaoqiang Liu

ICML 2026 · Image generation › Image restoration · approach: Diffusion-based universal restoration · 0 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 threshold tau to stop noise injection late in sampling (tau 3e-4 vs 0) (ablation)SubDAPS++ · FFHQ Gaussian deblurringFID: 65.14 -> 59.11Improved
raise noise-suppression threshold tau (ablation)SubDAPS++ · FFHQ Gaussian deblurringPSNR: 29.42 -> 28.95Hurt
add corrector step (ablation)SubDAPS++ · FFHQ high dynamic rangePSNR: 25.05 -> 25.52Improved
add corrector step (ablation)SubDAPS++ · ImageNet high dynamic rangePSNR: 23.57 -> 24.30Improved
use fine-tuned dynamic-resolution prior vs off-the-shelf Subspace DM prior (ablation)SubDAPS++ · FFHQ super-resolution 4xFID: 119.78 -> 64.61Improved
use ensemble of resolution-specific models instead of fine-tuned model (ablation)SubDAPS++ · FFHQ super-resolution 4xPSNR: 29.41 vs 29.34No clear effect
adapt DAPS to dynamic resolution (DAPS -> SubDAPS)DAPS · FFHQ inpainting and othersPSNR: 28.88 -> 28.89 (SR 4x)No clear effect
adapt DPS to dynamic resolution (DPS -> SubDPS)DPS · FFHQ super-resolution 4xPSNR: 24.00 -> 23.74Hurt
run SubDAPS++ at full dimension (SubDAPS-F++) instead of dynamic resolution (ablation)SubDAPS++ · six restoration tasksPSNR/FIDNo clear effect
use dynamic resolution (SubDAPS++ vs SubDAPS-F++) for speed (ablation)SubDAPS++ · ImageNet high dynamic rangetime (s): 9.2 -> 7.4Improved
operate SubDPS at 64x64 resolution vs DPSSubDPS · ImageNet HDRmemoryImproved
add SubDAPS++ improvements (CG, late deterministic, corrector) on SubDAPSSubDAPS · FFHQ motion deblurringFID: 82.45 -> 60.43Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

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

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