Multi-Head Self-Attention Mechanism for Enhanced Deepfake Detection
Sarthak Kulkarni, Dinesh Kumar Vishwakarma, Virender Ranga · 2024
Authentic detection of deepfakes can be vital to fighting misinformation, safeguarding security, and preventing financial and reputational damage. The more it acquires human nature, the more dangerous deepfake technology becomes. The ever-changing nature of deepfakes requires comprehensive measures allowing for the tackling of future risks—both fraud and misinformation.There is promise in the development of Vision Transformers (ViTs) for a new approach to detecting deepfakes. In contrast, ViTs consider images as sequences of patches applied with self-attention for modeling global relationships and do not use local receptive fields or any kind of hierarchical processing. This makes the ViTs much more robust, flexible, and scalable in learning long-range dependencies and contextual information. Which our new ViT-based model is supposed to aid in detecting built-in deep fakes and evaluate the model based on precision, recall, F1 Score.