Enhanced Parquet Edition: FLDCF: A Collaborative Framework for Forgery Localization and Detection in Satellite Imagery

This repository packages the three datasets released with FLDCF into Hugging Face Parquet configurations. It contains 13,126 images, with original PNG bytes embedded in Parquet image columns, native resolutions, supplied masks, and per-sample metadata columns.

Subsets

Config	Split	Rows	Image size	Pixel masks	Parquet files
fake-vaihingen	train	2,099	256 × 256	Yes	1
fake-vaihingen	test	525	256 × 256	Yes	1
fake-loveda	train	6,852	512 × 512	Yes	11
fake-loveda	test	1,720	512 × 512	Yes	3
splice-vaihingen	full	1,930	256 × 256	Not supplied	1

Columns

Shared columns:

• image: the original RGB image, decoded with the Hugging Face Image feature.
• label: image-level authenticity label, 0 = real, 1 = fake.
• manipulation: real, lama, repaint, or splice.
• sample_id, source_image_path: the original image path relative to its release directory.
• source_split: train or test for the inpainting datasets; unprovided for Splice-Vaihingen.
• filename_source_id: the area identifier or prefix extracted from the original filename.
• width, height: native image dimensions in pixels.
• has_mask: whether a pixel annotation is supplied.

fake-vaihingen and fake-loveda also include:

• mask: the original RGB pixel annotation; black (0, 0, 0) marks edited regions and white (255, 255, 255) marks background. Real images use an all-white mask.
• source_mask_path: the mask path relative to its release directory.

Metadata Files

Each configuration includes stats.json with split counts, manipulation counts, native dimensions, mask statistics where available, and Parquet shard paths and sizes.

The root inventory.json collects these statistics and records the total row count, shard count, and encoded byte sizes.

Citation

If you reference the FLDCF method or datasets, please cite the original paper:

@article{sui2024fldcf,
  title={FLDCF: A Collaborative Framework for Forgery Localization and Detection in Satellite Imagery},
  author={Sui, Jialu and Ma, Ding and Kuo, C.-C. Jay and Pun, Man-On},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  volume={62},
  pages={1--13},
  year={2024},
  doi={10.1109/TGRS.2024.3502035}
}

Dataset Export

This Parquet edition reorganizes the authors' released data and adds per-sample metadata. Please also reference this dataset repository when using the packaged version.

@misc{thanejoss_fldcf_2026,
  author={{ThaneJoss}},
  title={{FLDCF}},
  year={2026},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/ThaneJoss/FLDCF},
  note={Dataset available on Hugging Face}
}