davidqiuchao / nb2nb

This code implements the code of the paper ,"Neighbor2Neighbor:Self-Supervised Denoising From Single Noisy Images",in2021

Nb2Nb

This project aims to understand the idea of the paper ,"Neighbor2Neighbor:Self-Supervised Denoising From Single Noisy Images". Since this code is an unofficial implementation, some details may differ from the original description in the paper. To make it easier to understand the underlying theory, All the codes are written by Python and Tensorflow.

Sample Result

All the results are tested on SIDD Validation Dataset

Noisy Image	denoised Result

Update

Testing files, including trained model, are uploaded. The main testing file is "test.py", which can easily run by input the command .

python test.py -s saves -n nets.Unet -d dataDir -r resultDir

The "dataDir" specifies the testing data directory, and the "resultDir" is the path for saving result.

For rendering the '.mat' data, using the Simple Camera Pipline.

More training files will be uploaded as soon as possible. To be continue ...

nb2nb's Issues

Questions about the pretrained model and results.

Thanks for sharing it. I have two questions.

1. Do you train it with SIDD dataset? There are three different sizes: small, medium and full. How many epoches do you train for the final pretrained model? Would you mind offer the corresponding training code?
2. I test your model with some images from SIDD dataset. But I find some color distortions. Do your test results also suffer from it？Here is an example.
   Noisy:
   
   GT:
   
   Denoised:
   

It is obvious that the denoised image has a color distortion.
Looking forward to your replies! Thanks a lot.

requirments

holle,what is the tensorflow version? Can you tell us the enviroment requirements?thanks!

111