Advances in Media Forensics and Safety

The field of media forensics and safety is moving towards developing more sophisticated methods for detecting and localizing manipulated media, such as images and videos. Researchers are exploring new approaches, including context-aware contrastive learning and semantic-augment erasing, to improve the accuracy and generalizability of media forensic techniques. Additionally, there is a growing focus on ensuring the safety of generative models, particularly diffusion models, by developing methods to erase unacceptable concepts and prevent copyright infringement. Noteworthy papers include: RADAR, which proposes a novel approach for reliable identification of diffusion-based image manipulations, and SAGE, which introduces semantic-augment erasing to explore the boundaries of unsafe concept domains. These innovative techniques are advancing the field and have significant implications for real-world applications, such as image and video authentication, and safe content generation.

Sources

Towards Reliable Identification of Diffusion-based Image Manipulations

Unleashing the Potential of Consistency Learning for Detecting and Grounding Multi-Modal Media Manipulation

QualitEye: Public and Privacy-preserving Gaze Data Quality Verification

Context-aware TFL: A Universal Context-aware Contrastive Learning Framework for Temporal Forgery Localization

Do Concept Replacement Techniques Really Erase Unacceptable Concepts?

Real-Time Confidence Detection through Facial Expressions and Hand Gestures

SAGE: Exploring the Boundaries of Unsafe Concept Domain with Semantic-Augment Erasing

ME: Trigger Element Combination Backdoor Attack on Copyright Infringement

Built with on top of