AI Paper Atlas

Research map / Image generation / Image restoration

Single-Step Latent Diffusion for Underwater Image Restoration

Jiayi Wu, Tianfu Wang, Md Abu Bakr Siddique, Md Jahidul Islam, Cornelia M. Fermuller, Yiannis Aloimonos, Christopher A. Metzler

TPAMI 2025 · Image generation › Image restoration · approach: Diffusion-based universal restoration · 23 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 cross-latent decoder (ablation)SLURPP · Synthetic underwater (Osmosis simulated set)PSNR: 24.99 -> 25.66Improved
add reconstruction loss (enabled by single-step training) (ablation)SLURPP · Synthetic underwater (Osmosis simulated set)PSNR: 24.80 -> 25.66Improved
add inter-branch cross-attention (ablation)SLURPP · Synthetic underwater (Osmosis simulated set)PSNR: 25.06 -> 25.66Improved
single-step image-space training instead of iterative 50-step latent loss (ablation)SLURPP · Synthetic underwater (Osmosis simulated set)PSNR: 23.46 -> 25.66Improved
single-step training instead of iterative latent loss, both with cross-latent decoder (ablation)SLURPP · Synthetic underwater (Osmosis simulated set)PSNR: 24.62 -> 25.66Improved
add cross-latent decoder to iterative latent loss model (ablation)Iterative Latent Loss (50 steps) · Synthetic underwater (Osmosis simulated set)PSNR: 23.46 -> 24.62Improved
sample water parameters from real-world measurements instead of random values (ablation)SLURPP trained on Osmosis data · Synthetic underwater (Osmosis simulated set)PSNR: 24.87 -> 25.76Improved
use own terrestrial training data instead of Osmosis data (ablation)SLURPP with real-world water values · Synthetic underwater (Osmosis simulated set)PSNR: 25.76 -> 25.66No clear effect
random water parameters and Osmosis data still beat Osmosis baseline (ablation)SLURPP · Synthetic underwater (Osmosis simulated set)PSNR: 22.74 -> 24.87Improved
replace Osmosis with SLURPP for medium prediction (transmission)SLURPP · Simulated NYU underwaterPSNR: 13.97 -> 24.69Improved
replace Osmosis with SLURPP for medium prediction (backscatter)SLURPP · Simulated NYU underwaterPSNR: 23.08 -> 32.37Improved
single-step latent diffusion instead of iterative pixel-space diffusionSLURPP vs Osmosis · 512x512 inferenceruntime: more than 200 seconds -> 0.75 secondsImproved

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