Overview
DeeperForensics-1.0 is a large-scale benchmark dataset designed for real-world face forgery detection. It comprises 60,000 videos with a total of 17.6 million frames, making it significantly larger than existing datasets. The dataset incorporates extensive real-world perturbations, such as transmission errors and compression artifacts, to enhance its complexity and diversity. Fake videos are generated using an end-to-end face swapping framework (DF-VAE), ensuring high quality. DeeperForensics-1.0 includes a hidden test set with highly deceptive manipulated videos, challenging detection algorithms. It facilitates comprehensive evaluation of face forgery detection methods under diverse and realistic conditions, serving as a critical resource for advancing research in this field.