README.md

metadata

license: apache-2.0
language:
  - en
base_model:
  - prs-eth/marigold-depth-v1-0
  - stabilityai/stable-diffusion-2
tags:
  - underwater restoration
  - latent diffusion model
  - image analysis
  - computer vision
library_name: diffusers

SLURPP Single-Step Latent Diffusion Underwater Restoration

This is a model card for the slurpp model for underwater image restoration and water medium estimation from a single image. The model is fine-tuned from both stable-diffusion-2 model and marigold-depth-v1-0 model.

This model is based on the paper:

Jiayi Wu, Tianfu Wang, Md Abu Bakr Siddique, Md Jahidul Islam, Cornelia Fermuller, Yiannis Aloimonos, Christopher A. Metzler "Single-Step Latent Diffusion Underwater Restoration," IEEE Transactions on Pattern Analysis and Machine Intelligence, 2025

Using the model

Please follow the inference script in the github repo.

Model Details

• Developed by: Jiayi Wu, Tianfu Wang, Md Abu Bakr Siddique, Md Jahidul Islam, Cornelia Fermuller, Yiannis Aloimonos, Christopher A. Metzler.
• Model type: Diffusion-based single image underwater restoration and water medium prediction
• Language: English.
• License: Apache License License Version 2.0.
• Model Description: Given an input underwater image, our single-step latent diffusion method SLURPP jointly predicts the clear image, and the per-pixel underwater medium parameters, specifically the backscattering and transmission parameters.
  • Steps and scheduler: Only one step is required with this model using trailing DDIM scheduler
  • Outputs:
    • Clear: The predicted clear image
    • Transmission/Illuminaiton: The ammount of attenuation experienced by each object in the scene
    • Backscattering: The amount additive light from illumination (ambient sunlight or artificial lights) that is scattered by the water volume
• Resources for more information: Project Website, Paper, Code.
• Cite as:

@article{wu2025single,
  title={Single-Step Latent Diffusion for Underwater Image Restoration},
  author={Wu, Jiayi and Wang, Tianfu and Siddique, Md Abu Bakr and Islam, Md Jahidul and Fermuller, Cornelia and Aloimonos, Yiannis and Metzler, Christopher A},
  journal={arXiv preprint arXiv:2507.07878},
  year={2025}
}