English (United Kingdom)

https://curated-unify.zendy.io/wp-json/zendy-region/v1/featured_content/oa?rat=en

https://curated-unify.zendy.io/wp-json/zendy-region/v1/highlighted_journal/

Global: Zendy Plus annual

Presents the access of premium content as premium feature

Premium Content

Presents the keyphrase highlighting as premium feature

Keyphrase Highlighting

Presents the summarisation as premium feature

Summarisation

Insights

Presents the pdf analysis as premium feature

PDF Analysis

Presents the zaia usage as premium feature

ZAIA

Global: Zendy Plus monthly

Global: Zendy Tools annual

Global: Zendy Tools monthly

Global: Zendy Free Plan

Deepfakes, malicious visual contents created by generative models, pose anincreasingly harmful threat to society. To proactively mitigate deepfakedamages, recent studies have employed adversarial perturbation to disruptdeepfake model outputs. However, previous approaches primarily focus ongenerating distorted outputs based on only predetermined target attributes,leading to a lack of robustness in real-world scenarios where target attributesare unknown. Additionally, the transferability of perturbations between twoprominent generative models, Generative Adversarial Networks (GANs) andDiffusion Models, remains unexplored. In this paper, we emphasize theimportance of target attribute-transferability and model-transferability forachieving robust deepfake disruption. To address this challenge, we propose asimple yet effective disruption method called Latent Ensemble ATtack (LEAT),which attacks the independent latent encoding process. By disrupting the latentencoding process, it generates distorted output images in subsequent generationprocesses, regardless of the given target attributes. This targetattribute-agnostic attack ensures robust disruption even when the targetattributes are unknown. Additionally, we introduce a Normalized GradientEnsemble strategy that effectively aggregates gradients for iterative gradientattacks, enabling simultaneous attacks on various types of deepfake models,involving both GAN-based and Diffusion-based models. Moreover, we demonstratethe insufficiency of evaluating disruption quality solely based on pixel-leveldifferences. As a result, we propose an alternative protocol forcomprehensively evaluating the success of defense. Extensive experimentsconfirm the efficacy of our method in disrupting deepfakes in real-worldscenarios, reporting a higher defense success rate compared to previousmethods.

LEAT: Towards Robust Deepfake Disruption in Real-World Scenarios via Latent Ensemble Attack