Generalized Deepfake Attribution

Authors: Sowdagar Mahammad Shahid, Sudev Kumar Padhi, Umesh Kashyap, Sk. Subidh Ali

Published: 2024-06-26 12:04:09+00:00

AI Summary

This paper introduces a Generalized Deepfake Attribution Network (GDA-Net) designed to attribute fake images to their source GAN architectures. GDA-Net overcomes limitations of existing methods by accurately identifying GAN architectures even when images are generated from retrained (cross-seed) or fine-tuned models. Extensive experiments confirm its superior effectiveness compared to current approaches.

Abstract

The landscape of fake media creation changed with the introduction of Generative Adversarial Networks (GAN s). Fake media creation has been on the rise with the rapid advances in generation technology, leading to new challenges in Detecting fake media. A fundamental characteristic of GAN s is their sensitivity to parameter initialization, known as seeds. Each distinct seed utilized during training leads to the creation of unique model instances, resulting in divergent image outputs despite employing the same architecture. This means that even if we have one GAN architecture, it can produce countless variations of GAN models depending on the seed used. Existing methods for attributing deepfakes work well only if they have seen the specific GAN model during training. If the GAN architectures are retrained with a different seed, these methods struggle to attribute the fakes. This seed dependency issue made it difficult to attribute deepfakes with existing methods. We proposed a generalized deepfake attribution network (GDA-N et) to attribute fake images to their respective GAN architectures, even if they are generated from a retrained version of the GAN architecture with a different seed (cross-seed) or from the fine-tuned version of the existing GAN model. Extensive experiments on cross-seed and fine-tuned data of GAN models show that our method is highly effective compared to existing methods. We have provided the source code to validate our results.

Key findings

GDA-Net, particularly the Denoiser-FEN variant, achieved significantly high accuracy (99.2%) on cross-seed data and demonstrated robustness against fine-tuned GAN models, outperforming state-of-the-art methods. The approach successfully extracts architecture-level traces using supervised contrastive learning, allowing generalized attribution across different GAN model instances.

Approach

The GDA-Net employs a Feature Extraction Network (FEN) trained with supervised contrastive learning to extract seed-independent, architecture-specific features. A denoising autoencoder is used to generate residual images, fed into the FEN, to minimize content dependency. These extracted features then train a multi-class classification network for final attribution.

Datasets

CelebA

Model(s)

UNKNOWN

Author countries

India

← Previous