Hierarchical Deep Fusion Framework for Multi-dimensional Facial Forgery Detection -- The 2024 Global Deepfake Image Detection Challenge

In recent years, advancements in generative artificial intelligence have led to the creation of highly realistic deepfakes, which present unprecedented threats to information integrity and personal security. The ability to manipulate facial images and videos through techniques like face swapping, attribute editing, and full synthesis necessitates the development of advanced detection systems. The Multi-dimensional Facial Forgery Detection challenge aims to address this issue by providing a comprehensive and diverse dataset, MultiFFDI, which includes a wide variety of forgery types generated by over 50 different methods. Traditional detection methods often struggle to generalize across unseen forgery techniques. To overcome this limitation, ensemble learning has emerged as a powerful strategy, combining the predictions of multiple models to improve overall robustness and accuracy. However, simply averaging outputs may not be optimal.