DCD-Net

This is the code for Iterative denoiser and noise estimator for self-supervised image denoising , ICCV 2023, by Yunhao Zou, Chenggang Yan and Ying Fu.

Introduction

In this work, we propose a Denoise-Corrupt-Denoise training pipeline (DCD-Net) for self-supervised image denoising. By iteratively updating the denoiser and noise estimator, DCD-Net achieves promising results on widely used image denoising benchmarks.

Data Preparation

• Download SIDD-Medium Dataset
• Put the datasets in folder ./test_dir, you can either use the SIDD validation set or testing set

Evaluation

• We provide the pretrained model of our DCD-Net in ./pretrained/dcd.pth, we also provide our reimplementation of other denoising methods in folder ./pretrained, including fully supervised baseline (n2c), noise2noise (n2n), and noise2void (n2v)
• Run the following script for evaluation

bash test.sh

Citation

If you find this work useful for your research, please cite:

@inproceedings{zou2023iterative,
  title={Iterative denoiser and noise estimator for self-supervised image denoising},
  author={Zou, Yunhao and Yan, Chenggang and Fu, Ying},
  booktitle={Proceedings of the IEEE/CVF International Conference on Computer Vision},
  pages={13265--13274},
  year={2023}
}

Contact

If you have any problems, please feel free to contact me at [email protected]

Acknowlegment

The code borrows from Blind2Unblind, and Neighbor2Neighbor, please also cite their work