Skip to content

Latest commit

 

History

History
67 lines (45 loc) · 2.42 KB

README.md

File metadata and controls

67 lines (45 loc) · 2.42 KB
Raw

Adaptive Uncertainty Distribution in Deep Learning for Unsupervised Underwater Image Enhancement (Paper)

License Framework

The Pytorch Implementation of ''Adaptive Uncertainty Distribution in Deep Learning for Unsupervised Underwater Image Enhancement''.

Introduction

We present a fully trainable framework to enhance underwater images without ground truth. We demonstrate that our proposed model outperforms ten popular underwater image enhancement methods on seven common metrics, in both paired and unpaired settings.

Requirement

In this project, we use Linux host with a single NVidia GeForce RTX 2080 Ti GPU with 11 GB of memory, Python 3.7, Pytorch 1.9.1.

Running

Inference

Download the pretrained model pretrained model.

Check the model and image pathes in Inference.py, and then run:

python Inference.py

Training

To train the model, you need to download Underwater Dataset dataset.

Check the dataset path in Train.py, and then run:

python Train.py

Bibtex

If you find UDnet is useful in your research, please cite our paper:

@article{Saleh2022udnet,
    title = {{Adaptive Uncertainty Distribution in Deep Learning for Unsupervised Underwater Image Enhancement}},
    year = {2022},
    author = {Saleh, Alzayat and Sheaves, Marcus and Jerry, Dean and Azghadi, Mostafa Rahimi},
    month = {12},
    url = {https://arxiv.org/abs/2212.08983v1},
    doi = {10.48550/arxiv.2212.08983},
    arxivId = {2212.08983},
    keywords = {Convolutional Neural Net-works, Deep Learning, Index Terms-Computer Vision, Machine Learning, Underwater Image Enhancement, Variational Autoen-coder}
}

License

The code is made available for academic research purpose only. This project is open sourced under MIT license.

Credit

https://github.com/zhenqifu/PUIE-Net
https://github.com/deepxzy/USLN
https://github.com/xueleichen/PyTorch-Underwater-Image-Enhancement