Deepfake Detection - Unified Ensemble Model

A high-precision deepfake detection system powering a modern web application. This project utilizes a powerful Ensemble Architecture that combines three distinct deep learning models (CNN, LSTM, and ResNeXt) to achieve superior accuracy in distinguishing real images from AI-generated deepfakes.

---

🧠 The Architecture: Triple-Model Ensemble

Unlike simple detection systems, this project leverages three specialized models working in unison:

1. 🔍 CNN (Convolutional Neural Network)

• Role: Standard feature extraction.
• Strength: Excellent at detecting local visual artifacts and pattern irregularities typical in deepfakes.

2. ⏱️ LSTM (Long Short-Term Memory)

• Role: Spatial-Sequence Analysis.
• Architecture: A hybrid model that extracts features using CNN layers and then processes them as a sequence using LSTM.
• Strength: Catching inconsistencies across the "spatial flow" of an image, treating image segments like a sequence to find structural breaks.

3. 🕸️ ResNeXt (Custom Lightweight)

• Role: Grouped Feature Processing.
• Architecture: A custom implementation of ResNeXt blocks data into multiple paths (cardinality).
• Strength: Captures complex, multi-scale features that standard CNNs might miss, effectively seeing "more details" without a massive computational cost.

✨ The Ensemble (The "Meta-Learner")

The Ensemble Model aggregates predictions from all three networks. By combining their unique perspectives—spatial patterns (CNN), sequential structure (LSTM), and multi-path features (ResNeXt)—it delivers a final confidence score that is more robust than any single model alone.

---

🚀 Features

• 🛡️ Multi-Model Security: Harder to fool thanks to the 3-model voting system.
• ⚡ FastAPI Backend: High-performance asynchronous API handling requests.
• 🎨 Modern Frontend: Beautiful, dark-mode UI with drag-and-drop support.
• 📊 Detailed Analytics: Returns confidence scores for "Real" vs "Fake" probabilities.
• 🏎️ GPU Acceleration: Automatic GPU detection for fast inference.

---

🛠️ Project Structure

DeepFakeDetection/
├── app.py                    # FastAPI Backend (Inference Server)
├── requirements.txt          # Project Dependencies
├── frontend/                 # Modern Web Interface
│   ├── index.html           
│   ├── styles.css           
│   └── script.js            
└── src/                      # Model Source Code
    ├── model_ensemble.py     # 🧠 The implementation of the Ensemble logic
    ├── model_cnn.py          # CNN Architecture
    ├── model_lstm.py         # LSTM Hybird Architecture
    ├── model_resnext.py      # ResNeXt Architecture
    └── train_ensemble.py     # Script to train all models together

---

⚡ Quick Start

1. Install Dependencies

pip install -r requirements.txt

2. Run the System

Start the backend server:

uvicorn app:app --reload

Access the API at http://localhost:8000

3. Launch Frontend

Open frontend/index.html in your browser.

---

🔧 Training the Ensemble

To train the full system from scratch, use the provided training scripts in the src/ directory.

1. Prepare Dataset: Place your Real/Fake dataset in data/Dataset/.
2. Train Models:

   # Train the complete ensemble
   python src/train_ensemble.py

   This script will train all individual models and evaluate the ensemble performance.

---

⚠️ Note on Model Files

Due to GitHub file size limits, the trained .h5 model files (which can imply huge sizes for an ensemble) are not included in this repository.

• You interpret the code to understand the architecture.
• Use the training scripts to generate your own fresh models!

---

📜 License

MIT License - Open for any use.