TST2K: A High-Quality Benchmark for Tone Style Transfer

TST2K is a benchmark for reference-based tone style transfer in photo retouching. It contains 2,000 image triplets built from the public PPR10K and MIT-Adobe FiveK research datasets.

Each triplet includes a content image, a reference image, and a ground-truth target image. The goal is to match the color and tone of the reference image while keeping the content and structure of the original image. TST2K covers a range of common photography scenes, including portraits, food, landscapes, night scenes, and lifestyle photos.

This dataset is released with our paper "Towards In-Context Tone Style Transfer with A Large-Scale Triplet Dataset."

• Project page: ICTone
• Paper: arXiv:2604.16114

File Structure

TST2K/
|-- 0000/
|   |-- content.png
|   |-- reference.png
|   `-- gt.png
|-- 0001/
|   |-- content.png
|   |-- reference.png
|   `-- gt.png
|-- ...
`-- 1999/
    |-- content.png
    |-- reference.png
    `-- gt.png

• content.png: the input photo whose content and structure should be preserved.
• reference.png: the reference photo that provides the target color and tone.
• gt.png: the ground-truth retouched result.

The content and ground-truth images have the same resolution. The reference image may have a different size or aspect ratio.

Data Sources and License

TST2K is derived from two public research datasets:

• PPR10K: released for non-commercial research only. Its terms also restrict commercial use of the images and derived data.
• MIT-Adobe FiveK: also known as MIT5K. Its images are covered by LicenseAdobe.txt or LicenseAdobeMIT.txt. Both licenses allow research use but prohibit use for commercial gain or monetary compensation.

Because TST2K is derived from these datasets, it is available for non-commercial academic and research use only. Commercial training, fine-tuning, product development, evaluation of commercial systems, resale, and use in commercial products or services are not allowed.

Citation

If you use TST2K or the tone style scorer in your research, please cite:

@misc{deng2026incontexttonestyletransfer,
  title         = {Towards In-Context Tone Style Transfer with A Large-Scale Triplet Dataset},
  author        = {Yuhai Deng and Huimin She and Wei Shen and Meng Li and Ruoxi Wu and Lunxi Yuan and Xiang Li},
  year          = {2026},
  eprint        = {2604.16114},
  archivePrefix = {arXiv},
  primaryClass  = {cs.CV}
}