PanScale / README.md
kecao's picture
Add files using upload-large-folder tool
0d7abda verified
metadata
pretty_name: PanScale
license: cc-by-4.0
task_categories:
  - image-to-image
tags:
  - remote-sensing
  - pansharpening
  - multispectral
  - panchromatic
size_categories:
  - 1K<n<10K

PanScale

arXiv GitHub Hugging Face Dataset

Dataset Summary

PanScale is a remote-sensing pansharpening dataset with paired multispectral (ms) and panchromatic (pan) TIFF images for cross-scale evaluation.

  • Total pairs: 7,559
  • Disk size: ~6.9 GB
  • Format: 8-bit TIFF

Supported Tasks

  • Image Fusion (Cross-scale Pansharpening)

Subsets at a Glance

Subset Splits # Pairs MS size PAN size PAN/MS scale
jilin train200, test200/400/800 1,157 200-800 200-800 1.0
landsat train256, test200/400/800/1600 2,868 200-1600 200-1600 1.0
skysat train200, test200/400/800/1600 2,722 200-1600 200-1600 1.0
fjilin test800/1200/1600/2000 484 200-500 800-2000 4.0
flandsat test800/1200/1600/2000 192 400-1000 800-2000 2.0
fskysat test800/1200/1600/2000 136 320-800 800-2000 2.5

Data Fields

  • ms: 4-channel multispectral image (.tif)
  • pan: 1-channel panchromatic image (.tif)
  • id: filename stem (integer)

Intended Use

  • Pansharpening training and evaluation
  • Cross-scale robustness analysis

License

This dataset is released under CC BY 4.0.

Citation

Citation information will be added after the paper is publicly available.

@article{cao2026cross,
  title={Cross-Scale Pansharpening via ScaleFormer and the PanScale Benchmark},
  author={Cao, Ke and He, Xuanhua and Li, Xueheng and Zhu, Lingting and Wang, Yingying and Ma, Ao and Zhang, Zhanjie and Zhou, Man and Xie, Chengjun and Zhang, Jie},
  journal={arXiv preprint arXiv:2603.00543},
  year={2026}
}