image-detector

Fake Image Detector

A deep learning-powered web app that detects whether an image of a human face is AI-generated (fake) or real.

This project combines a custom PyTorch convolutional neural network (CNN) with a full-stack Flask + React frontend. Users can upload or select images, receive real-time predictions with confidence scores, and visualize saliency maps that explain the model's decision.

💻 Tech Stack

• Frontend: React, SCSS
• Backend: Flask, PyTorch, Captum (for saliency maps)
• Model: Custom CNN trained to classify real vs AI-generated faces (87% accuracy)
• Other: Numpy, Matplotlib, Pillow

📸 Screenshots + Live Demo

---

🎯 Features

• Upload or drag-and-drop image input
• Real-time classification of image as REAL or FAKE
• Saliency map generation to visualize which image regions influenced the model
• Confidence scores for each prediction
• Sample image gallery (Ed Sheeran, Michelle Obama, etc.)
• Light/dark mode toggle
• Mobile responsive UI

---

🧠 Model

A lightweight CNN was trained on real and GAN-generated face images using PyTorch. The model achieved 87% accuracy on the validation set. Saliency maps were generated with Captum to explain the model's attention.

Training steps included:

• Dataset preprocessing and 80/20 train-validation split
• Image normalization, resizing
• 20-epoch training loop using CrossEntropyLoss + Adam optimizer
• TorchVision for data loading and transforms

---

🚀 Run Locally

1. Clone the repo

   git clone https://github.com/yourusername/fake-image-detector
   cd fake-image-detector

2. Set up Python backend

   python3 -m venv venv
   source venv/bin/activate
   pip install -r requirements.txt
   python server/app.py

3. Set up React frontend

   cd client
   npm install
   npm start

📚 Future Improvements

• Upload history with timestamps
• Multiple model comparisons (e.g., VGG, ResNet)
• User authentication + saved predictions