Description

WildRF is a benchmark dataset for evaluating deepfake detection generalization to real-world, in-the-wild content extracted from social media. It is curated from several popular social networks to provide challenging deepfakes that reveal gaps between lab benchmarks and real-world performance.