HKLUT: Hundred-Kilobyte Lookup Tables

This repository serves as the official code release of the IJCAI24 paper: Hundred-Kilobyte Lookup Tables for Efficient Single-Image Super-Resolution

Binxiao Huang^1†, Jason Chun Lok Li^1†, Ran Jie¹, Boyu Li¹, Jiajun Zhou¹, Dahai Yu², Ngai Wong¹

^†Contributed Equally

¹The University of Hong Kong, ²TCL Corporate Research

Usage

Dependency

conda create -n hklut python=3.8
conda activate hklut
pip install -r requirements.txt

💾Dataset

Please follow the instructions below to download the corresponding datasets and place them in the specified folder structure.

Training set

The DIV2K dataset can be download from the offical website.

data/train/DIV2K/
            /HR/*.png
            /LR/{X2, X3, X4, X8}/*.png

Testing set

Please follow the instructions on BasicSR to download the SR benchmark datasets.

data/test/{Set5, Set14, B100, Urban100, Manga109}/
            /HR/*.png
            /LR_bicubic/{X2, X3, X4, X8}/*.png

🛠Quick Start

We have prepared bash scripts for each stage for convenient usage.

Training

./scripts/train.sh

Tranfer to LUTs

./scripts/transfer.sh

️Testing

./scripts/test.sh

📝Citation

If HKLUT has been beneficial to your research and applications, please acknowledge it by using this BibTeX citation:

@article{huang2023hundred,
  title={Hundred-Kilobyte Lookup Tables for Efficient Single-Image Super-Resolution},
  author={Huang, Binxiao and Li, Jason Chun Lok and Ran, Jie and Li, Boyu and Zhou, Jiajun and Yu, Dahai and Wong, Ngai},
  journal={arXiv preprint arXiv:2312.06101},
  year={2023}
}

🙏🏼Acknowledgements

Our code is built upon MuLUT and SPLUT. We sincerely appreciate their contributions to open-source.