Agamotto

MTCNN Architecture

Agamotto integrates advanced Machine Learning techniques, combining Convolutional Neural Networks (CNNs) for local feature extraction and  for capturing global relationships, delivering robust deepfake detection.

Advanced Preprocessing

Agamotto utilizes state-of-the-art deep learning libraries like MTCNN for precise face extraction, ensuring high-quality input for detection. It employs data augmentation techniques to increase dataset diversity, significantly improving the model’s ability to generalize across various scenarios

Effective Detection and Generalization

Optimized for detecting deepfakes across varied environments with a robust preprocessing pipeline and adaptable design for integrating new data and technologies.

Lightweight, Robust Model

We propose a lightweight model with 6.7 million parameters and a size of only 26 MB, suitable for classifying six different classes. Real videos and five different classes of deepfakes (FaceSwap, NeuralTextures, Face2Face, DeepfakeDetection, FaceShifter) are effectively identified.

Comprehensive Evaluation

Agamotto is evaluated with accuracy, AUC, and log loss metrics, ensuring the solution’s reliability and effectiveness across various use cases. The solution is designed with a modern UI for an intuitive, easy to use interface, facilitating effortless integration and deployment.

Scalable and Future-Ready Design

With support for containerization technologies like Docker, Agamotto facilitates flexible scaling and straightforward deployment across multiple platforms, including AWS and Azure.