Recorrupted-to-Recorrupted: Unsupervised Deep Learning for Image Denoising

This repository is an PyTorch implementation of the paper Recorrupted-to-Recorrupted: Unsupervised Deep Learning for Image Denoising. The network we adopted is DnCNN and our implementation is based on DnCNN-PyTorch. We give the author credit for the implementation of DnCNN.

1.Recorrupted-to-Recorrupted (R2R) Scheme

2.Experimental Results

Table 1. Quantitative comparison, in PSNR(dB)/SSIM, of different methods for AWGN removal on BSD68. The compared methods are categorized according to the type of training samples.

Table 2. Quantitative comparison, in PSNR(dB)/SSIM, of different non-learning and unsupervised methods for denoising real-world images from SIDD.

3.Implementations

(1).Dependencies

PyTorch(<0.4)
torchvision
OpenCV for Python
HDF5 for Python
tensorboardX (TensorBoard for PyTorch)

(2). Run Experiments on AWGN removal

Train R2R model for AWGN removal with noise level $\sigma =25$:

python3 train_AWGN.py --prepare_data --noiseL 25 --val_noiseL 25 --training R2R

(3). The code for real world image denoising on SIDD dataset can be found here.

4.Citation

@InProceedings{Pang_2021_CVPR,
    author    = {Pang, Tongyao and Zheng, Huan and Quan, Yuhui and Ji, Hui},
    title     = {Recorrupted-to-Recorrupted: Unsupervised Deep Learning for Image Denoising},
    booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
    month     = {June},
    year      = {2021},
    pages     = {2043-2052}
}

PangTongyao / Recorrupted-to-Recorrupted-Unsupervised-Deep-Learning-for-Image-Denoising