SIH 2024

Team Name: dy/dx

Overview

Deepfakes are synthetic media in which a person in an existing image or video is replaced with someone else's likeness. This project leverages the XceptionNet architecture to detect such deepfakes with high accuracy.

Problem Statement

• Problem Statement ID: 1683
• Problem Statement Title: Development of AI/ML based solution for detection of face-swap based deep fake videos

Demo Video

Watch the video

dydx_demo.mp4

Deepfake Detection Using XceptionNet CNN MODEL

This repository contains the implementation of the deepfake detection model using XceptionNet as presented in the paper: A. V and P. T. Joy, "Deepfake Detection Using XceptionNet," 2023 IEEE International Conference on Recent Advances in Systems Science and Engineering (RASSE), Kerala, India, 2023, pp. 1-5. Read the paper

Future Aspects

1. Browser Extension: Develop an extension to detect deepfake content directly in the browser and provide user feedback.

2. Social Media Links Support: Integrate functionality to analyze and detect deepfakes in videos shared via social media links.

3. Detection Using Audio Anomalies: Add audio analysis to identify anomalies that may indicate deepfake content.

4. Frame/Frames Responsible for DeepFake Flag: Implement a feature to identify specific frames responsible for deepfake detection, enhancing the granularity of analysis.

Model Files

You can download the trained models from the following link:

Download Model

Installation

Clone this repository:

git clone https://github.com/dhruuvd-1704/dy-dx.git
cd dy-dx

Create a Terminal for Backend and run:

uvicorn backend:app --reload

Create a new Terminal for Frontend and run:

streamlit run app.py

Tech Stack Used

WEB DEVELOPMENT

• Streamlit: For building interactive web interfaces for the deepfake detection system and blockchain integration.
• Web3: To interact with the Ethereum blockchain for querying and updating smart contracts.

BACKEND

• FastAPI: For handling video uploads, deepfake detection, and API responses efficiently.

Team:

Dhruv Desai
Atharva Humane
Niranjan More
Mithilesh Singh
Vaishnavi Hud
Kasturi Pawar