New deepfake detection method uses consistency margin prior to improve cross-dataset performance
A new arXiv paper proposes a consistency margin prior derived from world models to address the drop in deepfake detector performance when tested on videos from datasets different from those used in training.
Read the full story at arXiv cs.CV ↗
Timeline · 1 report
- 2026-11-18 05:00 · arXiv cs.CV
Can a World Model Predict a Deepfake Video? A Consistency Margin Prior for Deepfake Detection