Venus-DeFakerOne: Unified Fake Image Detection & Localization
Researchers have developed Venus-DeFakerOne, a unified model for detecting and localizing fake images, addressing the fragmentation in current fake image detection research. This new model integrates InternVL2 and SAM2 to simultaneously perform image-level detection and pixel-level forgery localization across various forgery types. DeFakerOne demonstrates state-of-the-art performance, outperforming existing methods on numerous benchmarks and showing robustness against advanced generative models. AI
IMPACT This unified approach to fake image detection could improve the robustness of digital content verification systems against increasingly sophisticated AI-generated forgeries.