Enhanced Parquet Edition: Columbia Uncompressed Image Splicing Detection Evaluation Dataset (ForgeryVCR 832x832)

This repository packages the Columbia evaluation assets released with ForgeryVCR into a Hugging Face Parquet configuration. It contains 363 images: 183 real and 180 spliced images, with original supplied PNG bytes, available masks, and per-sample metadata embedded in Parquet.

This is the 832 × 832 preprocessed version supplied by ForgeryVCR, preserving its PNG images and grayscale masks. The original Columbia release uses native-resolution TIFF/BMP images and color-coded edgemasks.

The original Columbia dataset does not publish a fixed train/validation/test partition. The full split contains all 363 images supplied in the two ForgeryVCR archives. Their upstream location under datasets/test records their use in the ForgeryVCR evaluation workflow.

Subsets

Config	Split	Rows	Image size	Pixel masks	Parquet files
columbia	full	363	832 × 832	180 spliced images	1

Columns

• image: the supplied RGB image, decoded with the Hugging Face Image feature.
• mask: the supplied grayscale annotation; black 0 marks background and white 255 marks edited regions. Intermediate gray values are preserved. Real images have no supplied mask and use null.
• label: image-level authenticity label, 0 = real, 1 = fake.
• manipulation: real or splice.
• sample_id, source_image_path: the image path relative to the extracted release directory.
• source_mask_path: the supplied mask path relative to the extracted release directory, or null.
• source_split: test, preserved from the upstream ForgeryVCR evaluation directory; this is not an official Columbia partition.
• filename_source_id: the original filename prefix before _sub_.
• width, height: supplied image dimensions in pixels.
• has_mask: whether a pixel annotation is supplied.

Metadata Files

The columbia configuration includes stats.json with sample counts, dimensions, mask statistics, and Parquet shard paths and sizes.

The root inventory.json collects these statistics and records total row counts and encoded byte sizes. provenance/ retains the supplied image/mask manifests and bounding-box JSON.

Citation

If you reference Columbia, please cite the original paper and dataset:

@inproceedings{hsu06crfcheck,
  author={Y.-F. Hsu and S.-F. Chang},
  title={Detecting Image Splicing Using Geometry Invariants and Camera Characteristics Consistency},
  booktitle={International Conference on Multimedia and Expo},
  year={2006},
  location={Toronto, Canada}
}

Dataset Export

This Parquet edition reorganizes the Columbia assets distributed with ForgeryVCR and adds per-sample metadata. Please also reference this dataset repository when using the packaged version.

@misc{thanejoss_columbia_2026,
  author={{ThaneJoss}},
  title={{Columbia}},
  year={2026},
  publisher={Hugging Face},
  url={https://huggingface.co/datasets/ThaneJoss/Columbia},
  note={Dataset available on Hugging Face}
}