An end-to-end image translation model with weight-map for color constancy

Place RCC dataset in this directory:

./datasets/ColorChecker_ Recommended

Place the mask image data in this directory:

./datasets/masks

The program needs to use tensorflow and opencv toolkit. It is recommended to install in the conda environment.

And here we create a new conda environment as follow:

Install the latest version of tensorflow-gpu using the following command:

conda install tensorflow-gpu

install opencv:

pip install opencv-python

Enter the folder where CCUnet.py and dataset are located, and use the following command under the created conda environment to start training:

python CCUnet.py fold1

Here fold1 means that it is the first fold of the dataset used for this training, and can be replaced by fold2 or fold3, which means the second or the third fold used for training.

After each epoch of training, the testing process is automatically executed.

The train folder is used to save the image results during the training process.

The test folder is used to save the image results during the testing process.

The model folder is used to save the network model which achieves the best results.

And records.txt is the log file, which records the results of the experiment.