Generated-image-detection

Research on utilizing GANs to discern between real and fake images

Overview

The generated-image-detection project is a research-driven initiative focused on using Generative Adversarial Networks (GANs) to detect AI-generated images. Our system leverages advanced deep learning techniques and repurposes GAN discriminators to differentiate between genuine and synthetic facial imagery.

Features

• Image Detection & Analysis:

  • Detects AI-generated images from real ones using custom-trained GAN discriminators.
  • Evaluates performance using metrics such as Accuracy, Precision, Recall, F1 Score, and ROC-AUC.

• Model Training & Evaluation:

  • Detailed research and training pipelines are provided in notebooks (e.g., GAN.ipynb, WGAN.ipynb, Final_Project_inference.ipynb, and Sample Inference.ipynb).
  • Fine-tuning of discriminators to enhance detection performance.

• Interactive Dashboard & Chatbot:

  • A React-based dashboard (see DashboardPage.jsx) displays prediction results, detection confidence, and performance metrics.
  • A dedicated chatbot component (see chatbot.jsx) offers project-specific assistance and answers queries regarding the image detection methodology.
  • Tour guide for users , for our project

Technology Stack

• Backend:
  • Express.js API integrating with LangChain’s ChatOpenAI for model queries.
  • Flask-based inference endpoint (using PyTorch) for real-time image analysis.
• Frontend:
  • React (Vite) with Tailwind CSS powering the interactive user interface.
• Deep Learning & Data Tools:
  • PyTorch, TensorFlow, and Torchsummary for model development.
  • Pandas, NumPy, and OpenCV for data preprocessing and augmentation.
  • Matplotlib for visualization of training and performance metrics.
• Other Tools:
  • dotenv, cookie-parser, cors for secure and efficient application management.

Getting Started

Prerequisites

• Node.js (v14+) and npm.
• Python 3.x with necessary deep learning libraries (see your model requirements in requirements.txt or environment.yml).
• GPU support is recommended for model training.

Installation

1. Clone the repository:

   git clone <repo-url>
   cd generated-image-detection

2. Install backend dependencies:

   npm install

3. Install and build the frontend:

   cd frontend/UI
   npm install
   npm run build

4. Configure Environment Variables:

   • Create a .env file in the project root with your configurations. For example:

     OPENAI_API_KEY=your_actual_api_key_here
     

Running the Project

• Development Mode:
  From the project root, run:

  npm run dev

• Production Mode:
  Build and run your production bundle:

  npm start

• Optional Inference with Flask:
  To run the separate Flask-based inference service:

  cd Flask-model
  python app.py

  Inference will be available at http://localhost:5001/api/detect-image.

Project Structure

• backend/
  Contains the Express.js API (index.js) that leverages LangChain’s ChatOpenAI for processing project-specific queries.

• frontend/UI/
  The React-based UI, including components like:

  • DashboardPage.jsx: The dashboard for visualizing detection results and performance metrics.

• Notebooks:
  Research and development notebooks (GAN.ipynb, WGAN.ipynb, Final_Project_inference.ipynb, Sample Inference.ipynb) that document model training, evaluation, and inference experiments.

• Flask-model/
  Contains the inference pipeline implemented in Python (using PyTorch) for real-time image detection.

Project Screenshots

Below are some screenshots of our project, showcasing different aspects of the AI-Image Detection system. All images are sourced from the frontend/UI/assets/ProjectSS folder.

Contributing

Contributions, issues, and feature requests are welcome! Feel free to open an issue or submit a pull request.

License

This project is licensed under the ISC License.

Acknowledgements

• The open-source community for libraries and frameworks that make this project possible.
• Researchers and practitioners in GANs and AI image detection for their valuable insights.
• OpenAI for providing powerful APIs and tools that influence our work.