A comprehensive platform named Veritrue.ai which takes in various input formats like video, image, audio and predicts whether it is a deepfake or a real one.
We used Python's Gradio to create the frontend dashboard that takes in different types of inputs and outputs whether the input is deepfake or real along with an explanation, a heatmap, and available EXIF or metadata.
We used
MTCNN (Multi-Task Cascaded Convolutional Neural Networks) to detect faces
Resnet and Inception algorithms to predict whether the images are deep fake or real
Melspectrogram Librosa algorithms to predict whether the audio is deep fake or real
This was done by the team TrioBoys (Arun Pranav A T, Adhithya S, Dhejan R) as a part of Tamil Nadu Police's CyberHackathon 3.0 which held at SRMIST Chennai
This project secured 4th place in the above mentioned hackathon