Selective Domain-Invariant Feature for Generalizable Deepfake Detection

View on arXiv ← Back to list

Authors: Yingxin Lai, Guoqing Yang Yifan He, Zhiming Luo, Shaozi Li

Published: 2024-03-19 13:09:19+00:00

AI Summary

The paper introduces SDIF, a novel framework for deepfake detection that improves generalizability by reducing sensitivity to irrelevant forgery details. It achieves this by fusing content and style features using a Farthest-Point Sampling strategy and a dynamic feature extraction module, while retaining domain-related features for better discrimination.

Abstract

With diverse presentation forgery methods emerging continually, detecting the authenticity of images has drawn growing attention. Although existing methods have achieved impressive accuracy in training dataset detection, they still perform poorly in the unseen domain and suffer from forgery of irrelevant information such as background and identity, affecting generalizability. To solve this problem, we proposed a novel framework Selective Domain-Invariant Feature (SDIF), which reduces the sensitivity to face forgery by fusing content features and styles. Specifically, we first use a Farthest-Point Sampling (FPS) training strategy to construct a task-relevant style sample representation space for fusing with content features. Then, we propose a dynamic feature extraction module to generate features with diverse styles to improve the performance and effectiveness of the feature extractor. Finally, a domain separation strategy is used to retain domain-related features to help distinguish between real and fake faces. Both qualitative and quantitative results in existing benchmarks and proposals demonstrate the effectiveness of our approach.

Key findings

SDIF outperforms state-of-the-art methods on in-domain and cross-domain deepfake detection tasks, achieving significantly higher accuracy and AUC scores. Ablation studies confirm the contribution of each module, particularly the importance of the diversity domain-aware module. The method shows robustness against different forgery types.

Approach

SDIF uses a three-module approach: 1) Farthest-Point Sampling generates a diverse style sample representation space; 2) a dynamic feature extraction module creates features with varying styles; 3) a domain separation strategy preserves domain-related features for improved discrimination between real and fake faces.

Datasets

FF++, Celeb-DF, WildDeepfake, DFDC

Model(s)

ResNet18 (pretrained on ImageNet), Dynamic Convolution, Adaptive Instance Normalization (AdaIN), Multilayer Perceptron (MLP)

Author countries

China

← Previous