Open Set Face Forgery Detection via Dual-Level Evidence Collection

TL;DR

This paper tackles Open Set Face Forgery Detection (OSFFD), where models must recognize novel fake categories while classifying real and known fakes. It introduces Dual-Level Evidential face forgery Detection (DLED), a framework that collects and fuses spatial (semantic via CLIP) and frequency (FFT-based) evidence to produce principled uncertainty estimates using Evidential Deep Learning. By reformulating OSFFD as an uncertainty estimation problem and applying a Dempster-Shafer-based fusion, DLED achieves state-of-the-art detection of unseen forgeries while maintaining competitive Real-vs-Fake performance on the DF40 dataset. The work demonstrates that combining spatial and frequency evidence with improved uncertainty calibration enables robust, interpretable open-set forgery detection suitable for real-world deployment.

Abstract

The proliferation of face forgeries has increasingly undermined confidence in the authenticity of online content. Given the rapid development of face forgery generation algorithms, new fake categories are likely to keep appearing, posing a major challenge to existing face forgery detection methods. Despite recent advances in face forgery detection, existing methods are typically limited to binary Real-vs-Fake classification or the identification of known fake categories, and are incapable of detecting the emergence of novel types of forgeries. In this work, we study the Open Set Face Forgery Detection (OSFFD) problem, which demands that the detection model recognize novel fake categories. We reformulate the OSFFD problem and address it through uncertainty estimation, enhancing its applicability to real-world scenarios. Specifically, we propose the Dual-Level Evidential face forgery Detection (DLED) approach, which collects and fuses category-specific evidence on the spatial and frequency levels to estimate prediction uncertainty. Extensive evaluations conducted across diverse experimental settings demonstrate that the proposed DLED method achieves state-of-the-art performance, outperforming various baseline models by an average of 20% in detecting forgeries from novel fake categories. Moreover, on the traditional Real-versus-Fake face forgery detection task, our DLED method concurrently exhibits competitive performance.

Figures (7)

• Figure 1: Comparison with existing settings. Different from DeepFake Detection (a) and Attribution (b), Open Set Face Forgery Detection (c) aims to identify whether a forgery originates from a novel fake category or not while simultaneously performing multiclass classification among real and known fake categories.
• Figure 2: Illustration for Fake Categories in OSFFD. Real faces and fake faces from the seen categories are used to train the model. Subsequently, the model is evaluated on test data that includes both seen classes and previously unseen categories. In the figure, the labels EFS, FR, and FS denote seen categories, whereas FE represents an unseen category.
• Figure 3: Overview of DLED. DLED collects and fuses evidence from both the spatial and frequency domains to estimate prediction uncertainty. Our improved uncertainty estimation $\hat{u}$ is applied to achieve better detection performance. $F_N$ represents the $N$-th fake category and $K$ is the total known class number. If the uncertainty for the given sample is larger than the computed threshold, its label will be reassigned to the novel fake category. In the evidence illustration, we present a demonstration of a three-class classification scenario ($K=3$).
• Figure 4: Visualization of Attention Map. Attention maps generated by the DLED model for the novel fake categories FS and FE are shown in subfigures (a) and (b), respectively.
• Figure 5: Visualization of Evidence Distribution. The evidence for seen fake categories FR and EFS is condensed in their corresponding corner with low uncertainty, while the evidence for novel fake categories FS and FE is sparse with higher uncertainty.