Dual Path Denoising Network for Real Photographic Noise

Yeong Il Jang, Yoonsik Kim, and Nam Ik Cho

[Paper]

Test code

Environments

Windows10 / Ubuntu 16.04
Tensorflow1.12
Python 3.6

Models

Pretrained models for real noise and AWGN can be downloaded from followed link:

[Models]

Test

python test.py 
--imagepath : Path of test images
--savepath : Path of denoised images [default : './results/'] 
--model : Model checkpoint [real / AWGN] [default : './models/DPDN']
--add_noise : Add AWGN to test image [default : False]
--sigma : Standard deviation of AWGN (in [0 to 255] scale) [default : 25]

Examples

To denoise real noisy images,

python test.py --imagepath ./RNI15/ --savepath ./RNIdenoised/ --model ./models/DPDN

To denoise synthetic noisy images (AWGN),

python test.py --imagepath ./CBSD68/ --savepath ./CBSDdenoised/ --model ./models/AWGN --add_noise --simga 25

Experimental Results

The denoised results for DPDN can be downloaded from followed links:

DND SIDDValidation RNI15

Visualized results

Citation

If you use the work released here for your research, please cite this paper:

@ARTICLE{9098102,
author={Y. I. {Jang} and Y. {Kim} and N. I. {Cho}},
journal={IEEE Signal Processing Letters},
title={Dual Path Denoising Network for Real Photographic Noise},
year={2020},
volume={27},
number={},
pages={860-864},}

jyicu/DPDN