OmniFake Benchmark for Unified Multimodal Misinformation Detection
July 31, 2026
The OmniFake benchmark contains 98K samples combining human-curated misinformation with AI-generated multimodal content to train detectors. The proposed UMFDet framework addresses both human and synthetic deception within a single unified model.
HOW THIS AFFECTS YOU
●
researcherYou can use this dataset to train models that generalize across human and synthetic fake content.
●
policyThis provides a more realistic framework for detecting complex, real-world misinformation campaigns.