Deep Joint Design of Color Filter Arrays and Demosaicing

This is our Keras implementation for reconstructing test images of known datasets using our trained models.

Prerequisites

Python 2 or 3
Keras (with any Theano or Tensorflow backend)

Getting Started

Installation

Install Keras along with your preferred backend. For Ubuntu, you can easily install by:

pip install tensorflow keras

Clone this repository and enter:

git clone https://github.com/bernardohenz/deep_joint_design_cfa_demosaicing.git
cd deep_joint_design_cfa_demosaicing

Downloading our trained models

Download trained models

python ./trained_models/download_trained_models.py

Install h5py in order to load the models:

pip install h5py

Replicating the results from our paper

Download test datasets

python ./datasets/download_datasets.py

Run the test script

python test.py

You can specify desired parameters like the following example

python test.py --datasets kodak --model our_4x4_noise --noise_std 4 --output_dir results_noise_std4

Parameter --datasets specifies the test dataset you want to evaluate ([all | kodak | mcm | hdrvdp | moire])

Parameter --model specifies the trained model to be loaded ([our_4x4_noise-free | our_4x4_noise | bayer ])

Parameter --noise_std specifies the noise std to be added to the original image (in the scale [0,255])

Parameter --output_dir specifies the folder where the reconstructions will be saved (the script will not save the reconstructions if this is not specified)

Running our models in images

Running the specified model on a particular image

python reconstruct_image.py --img_name datasets/kodak/kodim01.png --model our_4x4_noise-free --output_name out.png

Running the specified model on directory

python reconstruct_images_from_dir.py --dir datasets/kodak --model our_4x4_noise-free --output_dir results_kodak

Training code

The original code was in Keras v1. We have updated to work on Keras>2.0, please follow to this training repository for the training script.

Citation

If you use this code, please cite our paper

@article{HenzGastalOliveira_2018,
    author = {Bernardo Henz and Eduardo S. L. Gastal and Manuel M. Oliveira},
    title   = {Deep Joint Design of Color Filter Arrays and Demosaicing},
    journal = {Computer Graphics Forum},
    volume = {37},
    year    = {2018},
    }