Deepfake Detection Using CNN & RNN

Project Overview

This project detects face-swap-based deepfake videos using a deep learning model built with CNN (ResNext) and RNN (LSTM). The model analyzes video frames to determine whether a video is real or fake.

Features

• Deepfake Detection: Analyzes videos frame by frame.
• Confidence Score: Provides a probability of whether the video is fake.
• PDF Report Generation: Generates a detailed report on the analysis.
• User-Friendly Web Interface: Allows easy video uploads and model predictions.

---

1️⃣ Installation & Setup

Prerequisites:

• Python 3.8+
• pip (Python package manager)
• Virtual environment (optional but recommended)

Step 1: Clone the Repository

git clone https://github.com/SakshiAher-Codes/Detection-of-Face-Swap-based-Deep-Fake-Videos.git
cd Detection-of-Face-Swap-based-Deep-Fake-Videos

Step 2: Install Dependencies

Ensure you have all required dependencies installed by running:

pip install -r requirements.txt

Step 3: Run the Application Locally

streamlit run main.py

This will open the Streamlit web app in your browser.

---

2️⃣ Understanding the Deepfake Detection Process

1. Video Processing

• Extracts frames from the uploaded video.
• Preprocesses frames by resizing and normalizing them.

2. Model Prediction

• Each frame is passed through the trained ResNext + LSTM model.
• The model assigns a confidence score indicating if the frame is fake or real.

3. Report Generation

• Calculates an overall confidence score for the video.
• Generates a detailed PDF report with frame-wise analysis.

---

3️⃣ File Structure

📁 Detection-of-Face-Swap-based-Deep-Fake-Videos
│── 📄 main.py  # Streamlit app interface
│── 📄 deepfake_model.py  # Deepfake detection functions
│── 📄 requirements.txt  # Required Python packages
│── 📄 README.txt  # Project documentation

---

4️⃣ Troubleshooting

Error: ModuleNotFoundError: No module named 'gdown'

Fix: Install gdown manually:

pip install gdown

Error: NameError: name 'st' is not defined

Fix: Ensure import streamlit as st is present at the top of main.py.

Error: Yowza, that’s a big file (>25MB)!

Fix: Store the model on Google Drive and use:

import gdown
gdown.download("https://drive.google.com/uc?id=1UiYjPQBC-mZO4qETov4C6oFlnSNeSf2i", "deepfake_model.h5", quiet=False)

---

5️⃣ Required Software & Platforms

Software Required

• Python (Version 3.8+) – Core language for the project
• pip – Python package manager for installing dependencies
• Streamlit – To create the web interface
• TensorFlow / Keras – For deep learning model operations
• OpenCV – To process video frames
• ReportLab – To generate PDF reports
• gdown – To download the model from Google Drive

Platforms Required

• Operating System: Works on Windows, macOS, and Linux
• Python Environment: Recommended to use virtual environment for package isolation
• Google Drive: To store and access the trained deepfake model
• Streamlit Cloud (optional): If deploying the application online

---

📩 Contact

For any queries, feel free to reach out at biz.sakshiaher@gmail.com.

Happy Coding! 🚀