Unpaired Image Enhancement Featuring Reinforcement-Learning-Controlled Image Editing Software

This is the official implementation of the paper in AAAI2020. We provide the sample codes for training and testing and pretrained models on photo enhancement.

Requirements

• Python 3.6
• Chainer 7.0
• ChainerRL 0.6
• pyyaml
• matplotlib
• opencv-python
• numpy

Installation

1. Clone the repository.

   # Make sure to clone with --recursive
   git clone --recursive git@github.com:satoshi-kosugi/Unpaired-Image-Enhancement.git

   2. Download MIT-Adobe FiveK Dataset. We used Adobe Lightroom to decode the images to TIF format and resize the long side of the images to 512 px. Put original images and images retouched by Expert C in $ROOT/fivek_dataset/original/ and $ROOT/fivek_dataset/expertC/, respectively.

Training and testing

1. Train the framework.

   python train.py settings/photo_enhancement.yaml logs

2. Test the framework.

   python test.py settings/photo_enhancement.yaml logs --result_dir logs/20200115T223451.986831/96000_finish/test_results --load logs/20200115T223451.986831/96000_finish/

3. If you want to try the framework, you can use the demo code and the trained model.

   python demo.py settings/photo_enhancement.yaml --load_generator generator_weights_for_demo.npz --file_name $image_name

         The following windows will be displayed.

   

Reference

Our implementation is based on chainer_spiral. We would like to thank them.

Citation

If you find our research useful in your research, please consider citing:

@inproceedings{kosugi2020unpaired,
    title={Unpaired Image Enhancement Featuring Reinforcement-Learning-Controlled Image Editing Software},
    author={Kosugi, Satoshi and Yamasaki, Toshihiko},
    booktitle = {AAAI},
    year = {2020}
}