Diffusion Facial Forgery Detection
Paper • 2401.15859 • Published
Datasets:
DiFF — unfiltered research copy
This is an independently prepared copy, not an official DiFF author release.
Original encoded images are preserved without resizing, re-encoding, quality filtering, or content deduplication. All original files, including prompts and auxiliary assets, are stored in lossless raw parts. Supplied train/val/test membership is preserved as train/validation/test. Images without a verified source partition are kept in unassigned. The original directory token is retained in original_split; no random split is created.
| class_id | class_name | binary_label | binary_name | rows |
|---|---|---|---|---|
| 0 | real | 0 | real | 38567 |
| 1 | FS | 1 | fake | 75909 |
| 2 | FR | 1 | fake | 0 |
| 3 | EFS | 1 | fake | 269974 |
| 4 | FE | 1 | fake | 235769 |
| 5 | PA | 1 | fake | 0 |
T2I and I2I → EFS is this collection's taxonomy mapping. DiFF's FS and FE groups retain those labels. The DiFF paper groups DCFace under FS; this copy follows that benchmark grouping.
Full verification report: reports/fd9a9fc455bdb854/collection_report.json.
Raw complete: True. Schema complete: True. Published counts match: False. Decode failures retained: 22284.
A count match cannot establish exact release membership without an author-provided file manifest. The local report records every source file and leaf member.
Research citation: Harry Cheng, Yangyang Guo, Tianyi Wang, Liqiang Nie, and Mohan Kankanhalli. Diffusion Facial Forgery Detection. ACM Multimedia 2024, pp. 5939–5948. https://arxiv.org/abs/2401.15859
- Downloads last month
- 154