Training-Free Dual-System Framework for Talking Head Forgery Detection
A new research paper on arXiv proposes a Training-Free Dual-System (TFDS) framework to enhance self-supervised talking head forgery detection. The framework draws inspiration from the dual-system theory of human cognition, treating anomaly-like scores as System-1 and using lightweight refinement as System-2. The approach aims to improve discriminative capacity on hard cases without additional training, addressing generalization challenges in supervised detectors.
Key facts
- arXiv paper 2605.03390 proposes TFDS framework
- TFDS is training-free and enhances self-supervised detectors
- Inspired by dual-system theory of human cognition
- System-1 uses anomaly-like scores
- System-2 uses lightweight refinement
- Addresses generalization challenges in supervised detectors
- Focuses on score-based self-supervised detectors
- Improves discriminative ability on hard cases
Entities
Institutions
- arXiv