Rethinking Affine Transform for Efficient Image Enhancement: A Color Space Perspective

By Di Li, and Susanto Rahardja

Introduction

The codebase provides the official PyTorch implementation for the paper "Rethinking Affine Transform for Efficient Image Enhancement: A Color Space Perspective" (accepted by IEEE Transactions on Multimedia).

Dependencies

• Python 3 (Recommend to use Anaconda)
• PyTorch >= 1.0
• Opencv
• Imageio

Datasets

The paper use the FiveK and PPR10K datasets for experiments.

• FiveK : You can download the original FiveK dataset from the dataset homepage and then process images using Adobe Lightroom.

  • To generate the input images, in the Collections list, select the collection Input with Daylight WhiteBalance minus 1.5.
  • To generate the target images, in the Collections list, select the collection Experts/C.
  • All the images are converted to .PNG format.

• PPR10K: You can download the train_val_images_tif_360p PPR10K dataset from the dataset homepage.

  • We used the 360p resolution images as the training and testing resolution. One can find the files by downloading train_val_images_tif_360p dataset (around 91GB). We choose the expert_b results as groundtruth.
  • Following the original paper, we train with the first 8,875 files and validate with the last 2286 files.

The final directory structure is as follows.

./train/
    input/         # 8-bit sRGB train inputs
    output/        # 8-bit sRGB train groundtruth
./eval/
    input/         # 8-bit sRGB eval inputs
    output/        # 8-bit sRGB eval groundtruth

Train

• run

python train.py --train_data_dir ./train --eval_data_dir ./eval  --epochs=100 --cuda --gpu_ids 0

Test

• run

python test.py --eval_data_dir=./test --cuda --ckpt_path ./result/ckpts/epoch_100_iter_112500.pth --gpu_ids 0

Citation

If you find this repository useful, please kindly consider citing the following paper:

@ARTICLE{10812861,
  author={Li, Di and Rahardja, Susanto},
  journal={IEEE Transactions on Multimedia}, 
  title={Rethinking Affine Transform for Efficient Image Enhancement: A Color Space Perspective}, 
  year={2024},
  volume={},
  number={},
  pages={1-12},
  doi={10.1109/TMM.2024.3521826}}

License

Our project is licensed under a MIT License.