Face Forensics¶
Detection of facial image manipulation and synthesis. Methods exploit properties of human faces including 3D structure, appearance consistency, dynamics, and behavioral cues that are difficult for generative models to capture perfectly.
Key papers¶
- Swin Meets EfficientNet: Lightweight Architectures for GAN-Based Face Forensics — Hybrid CNN-Transformer architecture combining EfficientNet-B0 with Swin-Transformer for efficient detection of GAN-synthesized faces; emphasizes practical computational constraints and achieves 99% accuracy on synthetic face detection.
- FakeCatcher: Detection of Synthetic Portrait Videos using Biological Signals — Detection using biological signals (photoplethysmography); analyzes spatial coherence and temporal consistency of blood flow patterns.
- Exposing Deep Fakes Using Inconsistent Head Poses — 3D head pose inconsistency reveals facial landmark misalignment in deepfakes.
- FaceForensics++: Learning to Detect Manipulated Facial Images — benchmark dataset and evaluation of face manipulation detection methods.