AI Paper Atlas

Research map / Image generation / Image restoration

Physics-Driven Turbulence Image Restoration with Stochastic Refinement

Ajay Jaiswal, Xingguang Zhang, Stanley H. Chan, Zhangyang Wang

ICCV 2023 · Image generation › Image restoration · approach: Real-world restoration with unaligned data · 42 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 PiRN with differentiable physics simulator consistency loss instead of TurbNetturbulence restoration · PiRN-Syn testPSNR: 23.72 -> 25.40Improved
add stochastic refinement (conditional DDPM, 10-20 steps) on top of PiRN (ablation)PiRN · PiRN-Syn testNIQE: 6.5852 -> 5.5847Improved
add stochastic refinement on top of PiRN (ablation)PiRN · PiRN-Syn testPSNR: 25.40 -> 25.61No clear effect
PiRN vs TurbNet under strong turbulenceturbulence restoration · PiRN-Syn strong turbulencePSNR: 23.90 -> 25.02Improved
restore text images with PiRN before OCROCR on Turbulence Text · Turbulence TextAWDR: 0.623 -> 0.776Improved
restore video with PiRN-SR before MTCNN face detectionface detection on turbulence video · 400m/500m videosF1-score: 0.4331 -> 0.4648 (500m)Improved
condition the diffusion model on PiRN output instead of the degraded image (ablation)stochastic refinement · diffusion trainingtraining stabilityImproved
use stochastic AT-DDPM restoration instead of deterministic SwinIR before face detectionface detection · 500m videoF1-score: 0.4006 (SwinIR) -> 0.3997 (AT-DDPM)Hurt

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