CS 444 Deepfake Detector using CNNs and LSTMs

Source the Virtual Environment

To be able to run all the files inside this repository you need to create a virtual environment and install all required packages

First, create the virtual environment:

python3 -m venv venv

Then, enter the virtual environment:

source venv/bin/activate

Finally, install all the required packages:

Note: Make sure you have CMake installed

How to install CMake (macOS)

# Install Homebrew if not already installed
/bin/bash -c "$(curl -fsSL https://raw.githubusercontent.com/Homebrew/install/HEAD/install.sh)"

# Install CMake
brew install cmake

# Verify installation
cmake --version

How to install CMake (Windows)

# Using winget
winget install kitware.cmake

# Using Chocolatey
choco install cmake

How to install CMake (Linux/Ubuntu)

# Update package list
sudo apt update

# Install CMake
sudo apt install cmake

# Verify installation
cmake --version

Install Packages

pip install -r requirements.txt

Instructions for Flask Application

To run the Flask demo application to test on a video, change directory into the Flask folder:

cd Flask

Then, follow the instructions in the README.md file inside that directory to run the Flask application.

Instructions of Training Code

To run the Training code to train a model, change directory into the Train folder:

cd Train

Then, follow the instructions in the README.md file inside that directory to run the Training code.

Navigating through the Flask App

Step 1: Upload Video

• Select Video File: Click the "Choose File" button to select a video from your device.
• Submit for Detection: Click the "Upload" button to submit the video for analysis.

Step 2: View Results and Frames

• View Prediction Results: The application displays whether the video is classified as a deepfake along with a confidence score.
• Use Slider to View Frames: Navigate through the extracted frames using the slider below the results.

Features

• Interactive Frame Viewer: Allows users to scroll through video frames using a slider.
• Near Real-time Results: Infers and provides feedback from the model, with near real-time speed, on whether the uploaded video is a deepfake.

Reference(s)

Li, Y., et al. (2020). Celeb-DF: A Large-scale Challenging Dataset for DeepFake Forensics. IEEE CVPR.