/DparNet

This is the implementation of manuscript "Wide & deep learning for spatial & intensity adaptive image restoration".

Primary LanguagePythonBSD 3-Clause "New" or "Revised" LicenseBSD-3-Clause

DparNet: Degradation parameter assisted restoration network

Description

This is the implementation of manuscript "Wide & deep learning for spatial & intensity adaptive image restoration".

System requirements

Prerequisites

  • Ubuntu 18.04+ or Windows 10/11
  • NVIDIA GPU + CUDA (Geforce RTX 3090-Ti with 24GB memory, CUDA 11.1 was tested)

Installation

  • Python 3.7+
  • Pytorch 1.7.0+

Quick start

Dataset

We provided very small version of the constructed denoising and deturbulence dataset, deposited in ./denoising/data/ and ./deturbulence/data/.

Test and evaluation

  • Run python test.py in each sub-folder to run the pre-trained models. The restored results will be saved in ./results/ of each sub-folder.
  • Run python evaluate.py in each sub-folder to obtain the quantitative evaluations of the restored results. As very little data is provided here, the evaluation results will not be the same as in manuscript.

Training

  • Run python train.py to perform training with the default setting.

Please cite our paper in your publications if our work helps your research.

@article{wang2023wide,
  title={Wide \& deep learning for spatial \& intensity adaptive image restoration},
  author={Wang, Yadong and Bai, Xiangzhi},
  journal={arXiv preprint arXiv:2305.18708},
  year={2023}
}