🧠 Real vs. AI Image Detector

This project is a web-based tool that detects whether an image is real or AI-generated using a fine-tuned CVT (Convolutional Vision Transformer) model. It combines a Python backend with a simple HTML/JS frontend.

---

🚀 Features

• Frontend image gallery with click-to-analyze interaction
• Flask backend with a /serverTest endpoint
• Pretrained CVT-13 backbone with a custom classifier
• Automatically downloads and analyzes the image
• Returns classification (Real or AI Generated) with confidence score

---

📁 Project Structure

main/
│
├── models/
│   ├── model_epoch_XX.pth       # trained model weights
│   ├── custom_dataset.py
│   ├── model.py
│   ├── aidetector.py
│   ├── inference.py
│   ├── train.py
│   └── evaluate.py
│
├── server.py                    # Flask API server
│
└── website/
    ├── index.html
    ├── content_script.js
    └── styles.css               # (optional)

---

🛠️ Setup Instructions

1. 📦 Install Requirements

Make sure Python 3.9+ is installed. Then run:

pip install -r requirements.txt

---

2. 🧠 Train Your Model (optional)

If you don't already have model weights:

python3 main/train.py main/models/DATASET/train --total_epochs 10 --save_path main/models

This will generate .pth files like model_epoch_9.pth in the models/ folder.

---

3. ▶️ Start the Backend Server

Run the Flask server:

python3 main/server.py

It will run on:
http://127.0.0.1:5000

---

4. 🌐 Launch the Frontend

Navigate to the website/ folder and run a static file server:

cd website
python3 -m http.server

Then open your browser and go to:
http://localhost:8000

---

📸 Usage

1. Click on an image in the gallery.
2. The image will be sent to the backend.
3. The server returns whether it's Real or AI Generated, along with the confidence level.
4. The result is logged in the console or displayed (if UI supports it).

---

📌 Notes

• Make sure the model file is inside main/models/ and named something like model_epoch_24.pth
• The image URLs in the frontend must be accessible via full URLs (not "images/dog.jpg")
• This tool is for educational purposes — AI detection is an evolving field.

---

👩🏻‍💻 Author

Created by Anna Scribner, Michael Gilbert, Muskan Gupta, and Roger Tang during NSC Hack4Impact - we won 1st prize!