Practical Manipulation Model for Robust Deepfake Detection

TL;DR

This paper introduces a Practical Manipulation Model (PMM) that enhances deepfake detection robustness by simulating diverse forgeries and applying degradations, leading to improved performance and stability across datasets.

Contribution

The paper presents a new manipulation model that covers a broader range of forgeries and enhances detector robustness through realistic degradations, outperforming previous methods.

Findings

3.51% and 6.21% AUC improvements on DFDC and DFDCP datasets

Enhanced robustness to common image degradations

Significant performance gains over baseline models

Abstract

Modern deepfake detection models have achieved strong performance even on the challenging cross-dataset task. However, detection performance under non-ideal conditions remains very unstable, limiting success on some benchmark datasets and making it easy to circumvent detection. Inspired by the move to a more real-world degradation model in the area of image super-resolution, we have developed a Practical Manipulation Model (PMM) that covers a larger set of possible forgeries. We extend the space of pseudo-fakes by using Poisson blending, more diverse masks, generator artifacts, and distractors. Additionally, we improve the detectors' generality and robustness by adding strong degradations to the training images. We demonstrate that these changes not only significantly enhance the model's robustness to common image degradations but also improve performance on standard benchmark datasets. Specifically, we show clear increases of $3.51\%$ and $6.21\%$ AUC on the DFDC and DFDCP datasets, respectively, over the s-o-t-a LAA backbone. Furthermore, we highlight the lack of robustness in previous detectors and our improvements in this regard. Code can be found at https://github.com/BenediktHopf/PMM