Skip to content

Latest commit

 

History

History
86 lines (64 loc) · 2.6 KB

README.md

File metadata and controls

86 lines (64 loc) · 2.6 KB
Raw

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.

architecture

Requirements

  • Python 3.8
  • 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 [email protected]: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 (the Input/InputAsShotZeroed collection inside fivek.lrcat) in fivek_dataset/original/. And put retouched by Expert C (the Experts/C collection inside fivek.lrcat) in $ROOT/fivek_dataset/expertC/.

3. Install Python environment.

# Use Python 3.8 via pyenv for this repo
# If you don't have pyenv, see: https://github.com/pyenv/pyenv
pyenv local 3.8

# Use poetry to install dependencies
# If you don't have poetry, see: https://python-poetry.org/
poetry env use python3.8
poetry install

Training and testing

0. Activate Poetry virtual environment.

poetry shell

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.

demo

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}
}