Al\'em do Desempenho: Um Estudo da Confiabilidade de Detectores de Deepfakes

TL;DR

This paper introduces a comprehensive reliability assessment framework for deepfake detectors, evaluating transferability, robustness, interpretability, and efficiency, revealing both progress and limitations in current methods.

Contribution

It proposes a novel framework for assessing deepfake detectors beyond accuracy, addressing key aspects like robustness and interpretability, and analyzes five leading methods.

Findings

Significant progress in deepfake detection techniques.

Critical limitations identified in current methods.

Framework highlights areas for future improvement.

Abstract

Deepfakes are synthetic media generated by artificial intelligence, with positive applications in education and creativity, but also serious negative impacts such as fraud, misinformation, and privacy violations. Although detection techniques have advanced, comprehensive evaluation methods that go beyond classification performance remain lacking. This paper proposes a reliability assessment framework based on four pillars: transferability, robustness, interpretability, and computational efficiency. An analysis of five state-of-the-art methods revealed significant progress as well as critical limitations.