AdaForensics: Learning A Characteristic-aware Adaptive Deepfake Detector

Authors: Xiaoke Yang, Haixu Song, Xiangyu Lu, Shao-Lun Huang, Yueqi Duan

Published: 2026-08-03 12:41:35+00:00

AI Summary

AdaForensics is a characteristic-aware adaptive network designed for deepfake detection. It addresses the limitation of fixed deepfake detectors by learning both characteristic-agnostic and characteristic-specific embeddings, enabling dynamic adaptation to varying facial characteristics. This approach uses a two-branch HyperNetwork to adjust detector parameters based on input characteristics, leading to customized forgery detection.

Abstract

In this paper, we propose a characteristic-aware adaptive network named AdaForensics for deepfake detection. Most existing methods learn a fixed network to detect deepfakes based on carefully-designed network architectures. However, these methods employ the same deepfake detector for all the images despite of various facial characteristic, which fail to provide customized forgery detection for different individuals. To address this, our AdaForensics simultaneously learns characteristic-agnostic and characteristic-specific embeddings, where the detector dynamically adapts to varying faces with our designed hypernetwork on the fly. More specifically, our AdaForensics not only explores the shareable abstractions from various deepfake images, but also adapts the detector to the given characteristic at test time. To achieve this, we propose a two-branch HyperNetwork to learn an adaptive deepfake detector, which automatically adjusts the parameters based on characteristic of the input. Extensive experiments on widely-used datasets including FaceForensics, Celeb-DF and DFDC demonstrate our AdaForensics outperforms the state-of-the-art works.

Key findings

AdaForensics significantly outperforms state-of-the-art methods in both intra-dataset and cross-dataset evaluations, achieving the highest average AUC scores. The characteristic-conditioned knowledge, particularly the combination of characteristic-agnostic and characteristic-specific branches, was shown to be highly effective, leading to substantial improvements in deepfake detection accuracy across diverse datasets.

Approach

AdaForensics employs a two-branch HyperNetwork to create an adaptive deepfake detector. One branch learns characteristic-agnostic embeddings (shareable abstractions across deepfakes), while the other extracts characteristic-specific embeddings from the input face. These two sets of embeddings are then fused to generate dynamic weights for the primary detection network, which is based on Xception, allowing it to adapt to individual facial characteristics.

Datasets

FaceForensics++ (FF++), Celeb-DF (CDF), DeepfakeDetection (DFD), DFDCP, FaceShifter (Fsh), UADFV, DFDC

Model(s)

Xception, AdaFace (ResNet50 backbone for feature extraction)

Author countries

China

← Previous