FairReL: Deepfake Detection using Fairness-Aware Representation Learning
Computer Vision and Pattern Recognition
Summary
The authors found that current deepfake detectors tend to make more mistakes on faces from certain demographic groups, often wrongly labeling real faces as fake. Instead of broadly adjusting the entire model, they focused on two specific parts of the model that relate to these errors: detailed facial features and changes made during training. They introduced a new method called FairReL, which uses special techniques to reduce bias in these parts by separating and controlling the features linked to demographics. Their tests show that FairReL both improves detection accuracy on new datasets and makes the detector fairer across different groups.
Authors
Xiaoman Lu, Jiaqi Li, Shuntian Zheng, Huiping Chen, Yu Guan
Abstract
Although recent deepfake detectors achieve high overall accuracy, their errors remain unevenly distributed across demographic subgroups, with real faces from certain groups more often misclassified as fake. Existing fairness-aware detectors typically regularise the entire feature representation, without identifying or controlling the specific components that drive unfair predictions. Such coarse intervention can over-suppress useful forgery cues while leaving demographic structure in component-specific subspaces. To address this, we identify two subgroup-sensitive components: multi-scale spatial features, which encode local facial and forgery patterns, and fine-tuning-induced residual features, which adapt the backbone to the unfair training distribution. We propose FairReL, a fairness-aware representation-learning framework that targets both components with dedicated demographic supervision. FairReL uses an SVD-decomposed foundation-model backbone to isolate the fine-tuning-induced residual representation, and introduces two complementary losses. Group-Conditional Wavelet Decorrelation (GCWD) suppresses subgroup-imbalanced structure across spatial wavelet sub-bands, while Subspace-Localised Mean Alignment (SLMA) aligns subgroup means within each real/fake class in the residual representation. Experiments on FF++, Celeb-DF, DFD and DFDC show that, against the state-of-the-art fairness-aware detector, FairReL improves unseen-dataset AUC by 3.9% while reducing subgroup FPR disparity by 10.2%. Code is available at https://github.com/xiaoman89/FairReL .