/Noisy-As-Clean-TIP2020

PyTorch Code for "Noisy-As-Clean: Learning Self-supervised Denoising from Corrupted Image", TIP 2020.

Primary LanguagePython

Noisy-As-Clean

Install

  • python = 3.6
  • pytorch = 0.4
  • numpy
  • scipy
  • matplotlib
  • scikit-image

Training

To reproduce the results in the NAC paper:

  • Training and test for Set12 at the same time
 python nac_resnet_on_set12.py

Testing

The test is performed after the models are trained.

If you have pre-trained models, for AWGN denoising on Set12, you can modify the specific settings that need to test in test_nac_resnet_on_set12.py

 run test_nac_resnet_on_set12.py

Draw Figrues

The code for drawing the figures in ablation study is provided in the figures folder

 run plot_for_large_sigma.py

#Further comments: The code is mainly written by Yuan Huang.

Part of the code borrow from [Deep Image Prior]