VerifyAI

VerifyAI is an AI-powered tool designed to detect deepfake images, distinguishing between authentic photographs and images generated by AI models. This project utilizes a deep learning model integrated into a user-friendly web interface.

Features ✨

• AI vs. Real Image Classification: Classifies uploaded images as either "AUTHENTIC (REAL)" or "AI GENERATED (FAKE)".
• High Accuracy: Employs a ResNet-50 model fine-tuned on the CIFAKE dataset, achieving approximately 96% accuracy on the test set.
• Streamlit Dashboard: Provides an interactive web interface for easy image uploads and result visualization.
• (Optional) Flask API: Includes a basic Flask backend API endpoint (/analyze) for programmatic access (currently in api.py).

Model Details 🧠

• Architecture: Transfer Learning using ResNet-50 (pre-trained on ImageNet).
• Dataset: Trained on the CIFAKE dataset ([kaggle+2] https://www.kaggle.com/datasets/birdy654/cifake-real-and-ai-generated-synthetic-images), containing 100,000 training images (50k real, 50k AI) and 20,000 test images (10k real, 10k AI).
• Performance: Achieved ~95.93% accuracy on the independent test set.
  • Baseline Comparison: A simple CNN trained from scratch achieved ~86.96% accuracy but showed significant overfitting and bias.
• Input: Expects 128x128 pixel RGB images.
• Output: Binary classification (FAKE=0, REAL=1) with a confidence score.
• Explainability (Grad-CAM): Attempts to implement Grad-CAM were made but were unsuccessful due to persistent tooling/library conflicts with the nested model structure. This feature is currently omitted.

Technology Stack 💻

• Backend & Model: Python, TensorFlow/Keras
• Web Framework (Dashboard): Streamlit
• Web Framework (API): Flask (Optional, in api.py)
• Image Processing: Pillow, OpenCV (for dashboard visualization)
• Numerical Computing: NumPy
• Data Visualization: Matplotlib (used during development/evaluation)

Setup and Installation ⚙️

1. Clone the repository:

   git clone [https://github.com/Hussny-06/VerifyAI.git](https://github.com/Hussny-06/VerifyAI.git)
   cd VerifyAI

2. Create and activate a virtual environment:

   # Windows
   python -m venv venv
   .\venv\Scripts\activate
   
   # macOS/Linux
   python3 -m venv venv
   source venv/bin/activate

3. Install dependencies:

   pip install -r requirements.txt

4. (Optional - Git LFS): Git LFS was implemented for model files, ensure it's installed (git lfs install) and pull the large files (git lfs pull). .

Usage 🚀

There are two ways to run the application:

1. Streamlit Dashboard (Recommended Interface)

This provides the interactive web UI.

• Ensure your virtual environment is active.

• Run the following command from the project root directory:

  streamlit run app.py

• Open your web browser and navigate to the local URL provided (usually http://localhost:8501).

• Use the sidebar to navigate (currently only "Single Image" is implemented).

• Upload an image on the "Single Image" page to get a prediction.

2. Flask API (Optional Backend Endpoint)

This runs a simple API server. Note: Ensure you have renamed the original Flask app file to api.py as discussed.

• Ensure your virtual environment is active.

• Run the following command from the project root directory:

  python api.py

• The API will be available at http://localhost:5000. You can send POST requests to the /analyze endpoint with an image file attached (key: 'image').

Project Structure 📁

VerifyAI/
├── .git/                     # Git metadata
├── .venv/                    # Virtual environment (ignored by git)
├── frontend/                 # Optional Flask frontend (HTML/CSS/JS)
│   ├── index.html
│   ├── script.js
│   └── style.css
├── pages/                    # Streamlit pages
│   └── 1_Single_Image.py
├── .gitignore
├── api.py                    # Optional Flask API
├── app.py                    # Streamlit entrypoint
├── baseline_cnn_v1.h5        # Saved baseline model (reference)
├── LICENSE
├── README.md
├── requirements.txt
├── resnet50_v1.keras         # Trained model file
└── tl_history.npy            # Training history (numpy)

Future Work 🔮

• Implement Video Analysis: Extend the detection capabilities to analyze video files, potentially incorporating temporal analysis techniques (e.g., using CNN+LSTM or Transformers) to identify inconsistencies between frames.
• Enhance Streamlit Dashboard:
  • Implement the remaining pages (Batch Processing, Model Performance, Dataset Explorer, About Model) as outlined in the dashboard plan.
  • Add video upload and analysis functionality to the dashboard.
• Explainability: Re-attempt Grad-CAM integration using alternative methods or libraries if they become more stable, or explore other XAI techniques (like LIME or SHAP).
• Model Exploration: Experiment with other high-performing architectures like EfficientNet or Xception.
• Dataset Expansion: Augment the training data with images/videos from newer or more diverse generative models.
• Containerization: Package the application (either Flask API or Streamlit dashboard) using Docker for easier deployment and portability.
• Cloud Deployment: Deploy the application to a cloud platform (e.g., Streamlit Community Cloud, Heroku, AWS, Google Cloud).