Introduction

This is a project which just move the official project N2V to tensorflow. The N2V is referring to the approach for image denoising in the paper "Noise2Void - Learning Denoising from Single Noisy Images". Please cite them if you think this project can help you.

Preparation

for using this code, you have to do something else:

1. Install tensorflow-2.2.0

please refer to tensorflow for details.

2. Download the data

this project only implement the code to load the BSD68 dataset. And you can follow the two steps to prepare the dataset.

  1. open BSD68_reproducibility_data.zip in web viewer and download it to the folder "data".
  2. uncompress the "BSD68_reproducibility_data.zip" in "data" folder.

Training

then, you just input the following sentence to train it.

python train.py <gpu_id>

Testing

python predict.py <gpu_id> saved_weight_path

Result

final psnr on BSD68 is 27.91 dB while it is 27.71 dB in the paper.