metadata
title: Doc Protocol Data
description: >-
The 4 cross-domain test sets (T-SROIE, OSTF, TPIC-13, RTM). All samples in the
cross-domain test sets are cropped to 512 × 512 patches without additional
compression. The training set and the three in-domain test sets share the same
three forgery synthesis types: copy-move, splicing, and print-based edits. The
cross-domain test sets have more diverse forgery sources, including AIGC-based
text editing models and manual manipulation.
license: other
Doc Protocol Data
The 4 cross-domain test sets (T-SROIE, OSTF, TPIC-13, RTM). All samples in the cross-domain test sets are cropped to 512 × 512 patches without additional compression. The cross-domain test sets have more diverse forgery sources, including AIGC-based text editing models and manual manipulation.
Cross-domain test sets
| Dataset | Split | Domain | #Samples | Description |
|---|---|---|---|---|
| T-SROIE [Wang et al., 2022b] | Test | Cross-domain | 1,579 | Scanned receipts tampered using the AIGC text editing model SR-Net. |
| OSTF [Qu et al., 2025] | Test | Cross-domain | 3,046 | Natural scene text images tampered using eight different AIGC-based text editing models. |
| TPIC-13 [Wang et al., 2022a] | Test | Cross-domain | 589 | Naturally captured scene-text images tampered using the AIGC text editing model SR-Net. |
| RTM [Luo et al., 2025] | Test | Cross-domain | 3,444 | Document images with both synthetic and manual manipulations, covering copy-move, splicing, print, and erasure edits across diverse document types such as scanned forms. |
Citation
If you use this dataset, please cite:
@article{du2025forensichub,
title={ForensicHub: A unified benchmark \& codebase for all-domain fake image detection and localization},
author={Bo Du and Xuekang Zhu and Xiaochen Ma and Chenfan Qu and Kaiwen Feng and Zhe Yang and Chi-Man Pun and Jian Liu and Ji-Zhe Zhou},
journal={Advances in Neural Information Processing Systems},
year={2025}
}