Enhancing Generalization of Deepfake Detection through Latent Space Augmentation

Enlarging the forgery space through latent space augmentation can help models learn a more robust and generalizable decision boundary, mitigating overfitting to forgery-specific features.