Face Forgery Detection Advances

The field of face forgery detection is rapidly evolving, with a focus on developing more comprehensive and efficient detection methods. Recent research has centered around creating unified frameworks that can handle multiple related tasks, such as image and video classification, spatial localization, and temporal localization. These unified models have been shown to leverage multi-task learning, enabling the capture of generalized representations across tasks and facilitating fine-grained knowledge transfer. Another area of innovation is the incorporation of local facial information and the integration of Large Language Models to improve detection accuracy and interpretability. Additionally, there is a growing emphasis on open-set face forgery detection, which involves recognizing novel fake categories, and on developing methods that can detect forgeries in real-world scenarios with high efficiency and scalability. Notable papers in this area include: OmniFD, which introduces a unified framework for versatile face forgery detection, and M4-BLIP, which proposes a face-enhanced local analysis approach for multi-modal media manipulation detection. Further research has also explored the use of croppable signatures to detect deepfake images and the development of model-agnostic frameworks to address the domain-dominant issue in multi-domain face forgery detection.

Sources

OmniFD: A Unified Model for Versatile Face Forgery Detection

M4-BLIP: Advancing Multi-Modal Media Manipulation Detection through Face-Enhanced Local Analysis

JPEGs Just Got Snipped: Croppable Signatures Against Deepfake Images

Open Set Face Forgery Detection via Dual-Level Evidence Collection

A Sanity Check for Multi-In-Domain Face Forgery Detection in the Real World

Built with on top of