Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Comparative Analysis of Deepfake Detection Models: New Approaches and Perspectives

Matheus Martins Batista

TL;DR

Este estudo compara modelos de detecção de deepfake, com foco no GenConViT, frente às arquiteturas do DeepfakeBenchmark, usando WildDeepfake e DeepSpeak para avaliar generalização. O GenConViT combina AE/VAE com backbones ConvNeXt e Swin Transformer, alcançando acurácia de 95%+ e AUC superior em DeepSpeak após ajuste fino, demonstrando eficácia robusta contra falsificações em ambientes controlados. No entanto, a generalização para deepfakes mais realistas, como WildDeepfake, permanece desafiadora, exigindo dados mais diversos e técnicas multimodais. Os resultados destacam a viabilidade de arquiteturas híbridas para detecção de DF, bem como a importância de otimização de pipeline, uso de GPUs e benchmarking padronizado para progressos sustentados na área.

Abstract

The growing threat posed by deepfake videos, capable of manipulating realities and disseminating misinformation, drives the urgent need for effective detection methods. This work investigates and compares different approaches for identifying deepfakes, focusing on the GenConViT model and its performance relative to other architectures present in the DeepfakeBenchmark. To contextualize the research, the social and legal impacts of deepfakes are addressed, as well as the technical fundamentals of their creation and detection, including digital image processing, machine learning, and artificial neural networks, with emphasis on Convolutional Neural Networks (CNNs), Generative Adversarial Networks (GANs), and Transformers. The performance evaluation of the models was conducted using relevant metrics and new datasets established in the literature, such as WildDeep-fake and DeepSpeak, aiming to identify the most effective tools in the battle against misinformation and media manipulation. The obtained results indicated that GenConViT, after fine-tuning, exhibited superior performance in terms of accuracy (93.82%) and generalization capacity, surpassing other architectures in the DeepfakeBenchmark on the DeepSpeak dataset. This study contributes to the advancement of deepfake detection techniques, offering contributions to the development of more robust and effective solutions against the dissemination of false information.

Comparative Analysis of Deepfake Detection Models: New Approaches and Perspectives

TL;DR

Este estudo compara modelos de detecção de deepfake, com foco no GenConViT, frente às arquiteturas do DeepfakeBenchmark, usando WildDeepfake e DeepSpeak para avaliar generalização. O GenConViT combina AE/VAE com backbones ConvNeXt e Swin Transformer, alcançando acurácia de 95%+ e AUC superior em DeepSpeak após ajuste fino, demonstrando eficácia robusta contra falsificações em ambientes controlados. No entanto, a generalização para deepfakes mais realistas, como WildDeepfake, permanece desafiadora, exigindo dados mais diversos e técnicas multimodais. Os resultados destacam a viabilidade de arquiteturas híbridas para detecção de DF, bem como a importância de otimização de pipeline, uso de GPUs e benchmarking padronizado para progressos sustentados na área.

Abstract

The growing threat posed by deepfake videos, capable of manipulating realities and disseminating misinformation, drives the urgent need for effective detection methods. This work investigates and compares different approaches for identifying deepfakes, focusing on the GenConViT model and its performance relative to other architectures present in the DeepfakeBenchmark. To contextualize the research, the social and legal impacts of deepfakes are addressed, as well as the technical fundamentals of their creation and detection, including digital image processing, machine learning, and artificial neural networks, with emphasis on Convolutional Neural Networks (CNNs), Generative Adversarial Networks (GANs), and Transformers. The performance evaluation of the models was conducted using relevant metrics and new datasets established in the literature, such as WildDeep-fake and DeepSpeak, aiming to identify the most effective tools in the battle against misinformation and media manipulation. The obtained results indicated that GenConViT, after fine-tuning, exhibited superior performance in terms of accuracy (93.82%) and generalization capacity, surpassing other architectures in the DeepfakeBenchmark on the DeepSpeak dataset. This study contributes to the advancement of deepfake detection techniques, offering contributions to the development of more robust and effective solutions against the dissemination of false information.

Paper Structure

This paper contains 54 sections, 15 equations, 22 figures, 16 tables.

Table of Contents

  1. Introdução
  2. Referencial Teórico
  3. Deepfake (DF)
  4. Categorias de Deepfake
  5. Geração de Deepfakes
  6. Processamento de Imagens Digitais
  7. Recorte da Face
  8. DeepfFace
  9. Aprendizado de Máquina
  10. Redes Neurais Artificiais
  11. Perceptron
  12. Redes Neurais Multicamadas (MLP)
  13. Função de Ativação
  14. Retropropagação e Gradiente Descendente
  15. Funções de Perda
  16. ...and 39 more sections

Figures (22)

  • Figure 1: Exemplos de Deepfake de faces humanas por reconstituição, substituição, edição e síntese. Fonte: adaptado de mirsky2021creation.
  • Figure 2: Fluxo de Geração de Deepfake de faces humanas por reconstituição e substituição. Fonte: adaptado de mirsky2021creation.
  • Figure 3: Deepfake gerada com autoencoder. Fonte: adaptado de masood2021deepfakes.
  • Figure 4: Hierarquia de aprendizado de máquina. Fonte: Elaborado pelo autor.
  • Figure 5: Perceptron: representação dos sinais de entrada, pesos, soma ponderada e função de ativação. Fonte: bianchini2001.
  • ...and 17 more figures