Robust deepfake detector against deep image watermarking
Jian Yu, Xin Liu, Fengbiao Zan, Yanhan Peng · 2025
Read Robust deepfake detector against deep image watermarking by Jian Yu, Xin Liu, Fengbiao Zan, Yanhan Peng (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/pmc-12755805.txt.
Description
Deepfake detectors can be fooled in an unexpected way: images carrying hidden watermarks throw off models trained on clean pictures. Jian Yu and co-authors build a detector meant to hold up when deep watermarks are present. Their design adds a multi-scale attention block inside a standard image network and a feature dropout step that strips redundant detail the watermarks might exploit.
On test sets mixed with two watermarking schemes, the model matched rival detectors against the first scheme. Against the second, which stamps a signature into faces, it beat the baselines clearly. The margin was roughly ten points of accuracy when half the images were watermarked, and twenty points when all of them were.
Catalog record
| CREATOR | Jian Yu, Xin Liu, Fengbiao Zan, Yanhan Peng |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 31 KB |
| VERSION | 1.0 |
| LICENSE | CC BY 4.0 |
| RIGHTS | CC BY |
| SOURCE | Europe PMC |