Searching for the Fakes: Efficient Neural Architecture Search for General Face Forgery Detection

View on arXiv ← Back to list

Authors: Xiao Jin, Xin-Yue Mu, Jing Xu

Published: 2023-06-15 03:01:13+00:00

AI Summary

This paper presents an end-to-end neural architecture search (NAS) framework for deepfake detection. It automatically designs network architectures using a forgery-oriented search space and a novel performance estimation metric, leading to improved generalization across datasets.

Abstract

As the saying goes, seeing is believing. However, with the development of digital face editing tools, we can no longer trust what we can see. Although face forgery detection has made promising progress, most current methods are designed manually by human experts, which is labor-consuming. In this paper, we develop an end-to-end framework based on neural architecture search (NAS) for deepfake detection, which can automatically design network architectures without human intervention. First, a forgery-oriented search space is created to choose appropriate operations for this task. Second, we propose a novel performance estimation metric, which guides the search process to select more general models. The cross-dataset search is also considered to develop more general architectures. Eventually, we connect the cells in a cascaded pyramid way for final forgery classification. Compared with state-of-the-art networks artificially designed, our method achieves competitive performance in both in-dataset and cross-dataset scenarios.

Key findings

The proposed NAS-based method achieves competitive performance on in-dataset and cross-dataset evaluations compared to state-of-the-art manually designed networks. The use of a forgery-oriented search space and the novel performance estimation metric contribute to improved generalization and efficiency.

Approach

The authors use differentiable architecture search (DARTS) to automatically design a network architecture for deepfake detection. A forgery-oriented search space, incorporating central difference convolutions, is employed. A novel performance estimation metric guides the search for more generalizable models.

Datasets

FaceForensics++ (FF++) , Celeb-DF, WildDeepfake, DFDC-preview

Model(s)

A novel Cell Cascaded Pyramid Network (C2PN) architecture automatically designed through the proposed NAS framework. The framework utilizes central difference convolutions (CDC) within its search space.

Author countries

China

← Previous