A Brief Review for Compression and Transfer Learning Techniques in DeepFake Detection
Andreas Karathanasis, John Violos, Ioannis Kompatsiaris, Symeon Papadopoulos · 2025
Read A Brief Review for Compression and Transfer Learning Techniques in DeepFake Detection by Andreas Karathanasis, John Violos, Ioannis Kompatsiaris, Symeon Papadopoulos (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/arxiv-2504-21066.txt.
Description
Shrinking a deepfake detector to fit a phone or camera is hard, because small devices have little memory and weak processors. Andreas Karathanasis, John Violos, Ioannis Kompatsiaris, and Symeon Papadopoulos review the shrinking tricks in a 2025 paper. Their twin tools are compression, which trims a trained model, and transfer learning, which cuts the cost of training.
The team tested pruning, distillation, quantization, fine-tuning, and adapter methods across three image datasets. Both strategies held up well even at 90 percent compression, as long as test fakes came from the same generator as the training fakes. When the test set used unseen generators, accuracy slipped and a domain gap opened. The review maps where edge deployment of detectors is safe today and where it still breaks.
Catalog record
| CREATOR | Andreas Karathanasis, John Violos, Ioannis Kompatsiaris, Symeon Papadopoulos |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 17 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2504-21066