FRIDAY: Mitigating Unintentional Facial Identity in Deepfake Detectors Guided by Facial Recognizers

View on arXiv ← Back to list

Authors: Younhun Kim, Myung-Joon Kwon, Wonjun Lee, Changick Kim

Published: 2024-12-19 08:21:28+00:00

AI Summary

FRIDAY is a novel training method for deepfake detectors that mitigates the reliance on facial identity for detection. It uses a face recognizer alongside the deepfake detector during training, minimizing the similarity of their feature embeddings to force the detector to focus on synthetic artifacts rather than identity.

Abstract

Previous Deepfake detection methods perform well within their training domains, but their effectiveness diminishes significantly with new synthesis techniques. Recent studies have revealed that detection models often create decision boundaries based on facial identity rather than synthetic artifacts, resulting in poor performance on cross-domain datasets. To address this limitation, we propose Facial Recognition Identity Attenuation (FRIDAY), a novel training method that mitigates facial identity influence using a face recognizer. Specifically, we first train a face recognizer using the same backbone as the Deepfake detector. The recognizer is then frozen and employed during the detector's training to reduce facial identity information. This is achieved by feeding input images into both the recognizer and the detector, and minimizing the similarity of their feature embeddings through our Facial Identity Attenuating loss. This process encourages the detector to generate embeddings distinct from the recognizer, effectively reducing the impact of facial identity. Extensive experiments demonstrate that our approach significantly enhances detection performance on both in-domain and cross-domain datasets.

Key findings

FRIDAY significantly improved deepfake detection performance on both in-domain and cross-domain datasets compared to state-of-the-art methods. The optimal weighting parameter (λ) for the Facial Identity Attenuating loss was found to be 10. The method demonstrated robustness across various datasets and synthesis techniques.

Approach

FRIDAY trains a face recognizer with the same backbone as the deepfake detector. The recognizer is then frozen, and during deepfake detector training, a loss function minimizes the similarity between the embeddings from both models. This forces the deepfake detector to learn from artifacts instead of facial identity.

Datasets

FaceForensics++ (FF++), Celeb-DF v1 & v2, DeepfakeDetection (DFD)

Model(s)

EfficientNet-B3 (as backbone for both face recognizer and deepfake detector)

Author countries

South Korea

← Previous