CDnCNN-B for blind color image denoising - Tensorflow implementation

This model is trained for blind denoising at multiple noise levels ([0, 55]).

DnCNN paper (TIP 2017)

Results on CBSD68 dataset

Noise Level	CBM3D	CDnCNN-B
25	30.70	31.23
50	27.38	27.97

Getting Started

This code was tested with Python 2.7. It is highly recommended to use the GPU version of Tensorflow for fast training.

Prerequisites

natsort==5.4.1
numpy==1.14.5
tensorflow==1.10.0
Pillow==5.4.1

Training the network

First, 128x3000 patches are extracted from the CBSD432 images as follows:

python2 generate_patches_rgb_blind.py

Then train the network:

python2 main_blind.py --phase train

You can also control other paramaters such as batch size, number of epochs. More info inside main.py.

The checkpoints are saved in ./checkpoint folder. Denoised validation images are saved after each epoch in ./sample folder.

Tensorboard summaries

tensorboard --logdir=./logs

Testing using the trained network

To test the network for sigma=50:

python2 main_blind.py --phase test --sigma 50.0

Denoised images are saved in ./test folder.

Reference

• Code structure follows the repository DnCNN-Tensorflow of @crisb-DUT (dataset import and feeding, loading checkpoints etc.).