Skip to content

An AI-powered system that detects and explains deepfake and AI-generated images using CNNs, Vision Transformers, and frequency analysis.

Notifications You must be signed in to change notification settings

McMasterAI-Society/DeepFakeDetector

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

71 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

🧠 DeepFakeDetector – AI-Generated Image Detection

📖 Overview

DeepFakeDetector is a research and development project by members of the McMaster Artificial Intelligence Society focused on detecting AI-generated and deepfake images.

With the rapid advancement of image synthesis models like StyleGAN, Stable Diffusion, and DALL·E, distinguishing synthetic content from authentic media has become increasingly difficult. This project aims to address that challenge by training models capable of detecting subtle visual artifacts that reveal an image’s generative origin.

Our system will combine multiple approaches:

  • CNNs (Convolutional Neural Networks) for spatial feature detection.
  • Vision Transformers (ViTs) for global context understanding.
  • Frequency-domain analysis (FFT) to capture artifacts invisible in the pixel domain.

The final product will be a web-based dashboard that allows users to upload an image and receive:

  • A real vs. AI-generated classification score.
  • A visual explanation (saliency or attention map) showing why the image was flagged.

🎯 Project Goal

To create a reliable, interpretable, and accessible AI system that:

  1. Accurately detects AI-generated and deepfake images.
  2. Provides visual explanations to promote transparency and trust in AI systems.
  3. Offers a usable web platform for real-time image authenticity verification.
  4. Contributes open research and tools to the broader AI ethics and misinformation-detection community.

👥 Team

Name Role
Lukhsaan Elankumaran Project Lead
Md Nafieu Hossain Alif ML Engineer
Zuhair Qureshi ML Engineer
Krish Bhagirath Research & Modeling
Oriana Rueckert Data Engineer
Vihaan Singhal Data & Deployment Engineer
Andrew Wu Web Interface and Hosting

🧩 End Goal

By the end of the term, the project will deliver:

  • A trained hybrid CNN + ViT + FFT-based model for image authenticity detection.
  • Explainability tools such as Grad-CAM and attention heatmaps for visual insights.
  • A Next.js + FastAPI web app that performs real-time deepfake detection from user uploads.
  • Technical documentation and a short demo video presenting results and key findings.

About

An AI-powered system that detects and explains deepfake and AI-generated images using CNNs, Vision Transformers, and frequency analysis.

Resources

Contributing

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Contributors 9