Attack-Aware Deepfake Detection Under Counter-Forensic Manipulations
2169-3536
Abstract
The rapid proliferation of AI-generated and manipulated visual content requires forensic detectors that remain reliable after routine processing and deliberate counter-forensic modification. This paper presents an attack-aware deepfake detector that combines pretrained semantic features with fixed forensic residual responses through a lightweight fusion module and produces weakly supervised spatial evidence maps. During training, a worst-of-<inline-formula> <tex-math notation="LaTeX">$K$ </tex-math></inline-formula> strategy selects the most damaging transformation from sampled JPEG compression, resampling, regraining, seam processing, color adjustment, and transcoding operations. At inference, predictions from low-cost randomized views are aggregated to reduce sensitivity to processing variations while preserving localized evidence. On the held-out DeepFakeFace split, the proposed framework achieved a worst-case accuracy of 0.9917 using a single validation-selected operating point. Regrain was the most challenging condition and produced an expected calibration error of 0.0196. Five independently seeded runs, controlled component analyses, and matched ResNet-50 and Xception comparisons were used to assess stability and robustness. The experiments identify worst-of-<inline-formula> <tex-math notation="LaTeX">$K$ </tex-math></inline-formula> exposure as the main contributor to counter-forensic operating robustness, while the semantic stream provides the strongest standalone representation and the residual stream remains independently predictive. The resulting heatmaps indicate where forensic evidence is concentrated without requiring pixel-accurate manipulation masks. GitHub Repository: <uri>https://github.com/BRAIN-Lab-AI/Attack-Aware-Deepfake-Detection/</uri> Project Page: <uri>https://brain-lab-ai.github.io/Attack-Aware-Deepfake-Detection/</uri>