🛡️ SENTINEL

Spatial Intelligence Engine // Deepfake Forensics

---

📖 Overview

Sentinel is an advanced AI-powered forensic tool designed to safeguard digital integrity. Utilizing a hybrid neural network architecture (EfficientNet-B4 + LSTM), Sentinel analyzes video media frame-by-frame to detect manipulation artifacts and temporal inconsistencies typical of deepfakes.

"In an era of synthetic media, truth requires a Sentinel."

✨ Key Features

Feature	Description
🧠 Hybrid Intelligence	Combines Spatial analysis (EfficientNet) with Temporal analysis (LSTM) for high-accuracy detection.
🔍 Micro-Forensics	Performs precision checking on video sequences to detect pixel-level anomalies.
⚡ Real-Time Evidence	Extracts compromised frames with exact timestamps and confidence scores.
📼 Smart Playback	Click on any evidence frame to instantly jump to that exact moment in the video player.
🔒 Privacy First	All processing happens locally or within your Docker container. No external APIs required.

---

🛠️ Tech Stack

• Core:
• Backend:
• Frontend:
• Deployment:

---

🚀 Quick Start (Docker)

The fastest way to deploy Sentinel is via Docker. This ensures total isolation and consistent dependencies.

1. Build the Image

   docker build -t deepfake-detector .

2. Run the Container

   docker run -p 7860:7860 deepfake-detector

3. Access the System Open your browser and navigate to: 👉 http://localhost:7860

---

⚙️ Local Installation (Python)

If you prefer native execution (Mac/Linux/Windows), follow these steps:

1. Clone the Repository

git clone [https://github.com/suryawanshiheramb11/auraverse2.git](https://github.com/suryawanshiheramb11/auraverse2.git)
cd auraverse2

### 2. Setup Virtual Environment (Recommended)
```bash
# MacOS/Linux
python3 -m venv venv
source venv/bin/activate

# Windows
python -m venv venv
venv\Scripts\activate

3. Install Dependencies

pip install -r requirements.txt

Note: This will install PyTorch, TorchVision, FastAPI, Uvicorn, and other core libraries.

4. Verify Model

Ensure sentinel_model.pth is in the root directory. This file contains the trained weights for the SentinelHybrid model.

5. Run the Server

uvicorn server:app --reload --host 0.0.0.0 --port 8000

Note: In local mode, the app usually runs on port 8000.

6. Access the App

Open: http://localhost:8000

---

📂 Project Structure

├── backend/
│   ├── orchestrator.py    # Core analysis pipeline (Model Loading, Inference)
│   ├── schemas.py         # Data models (API Responses)
│   └── processing/        # Frame extraction utilities
├── static/
│   └── evidence/          # Extracted fake frames are saved here
├── Dockerfile             # Container configuration
├── server.py              # FastAPI server entry point
├── index.html             # Main UI (Sentinel V2.5 Blue Theme)
├── sentinel_client.js     # Frontend Logic & API Client
├── model_core.py          # PyTorch Model Architecture (EfficientNet+LSTM)
└── requirements.txt       # Python dependencies

🐛 Troubleshooting

"Failed to Fetch" Error

• Ensure the server is running.
• In Docker, make sure you mapped the ports correctly (-p 7860:7860).
• Check the terminal logs for backend errors.

Model Not Found

• The system looks for sentinel_model.pth in the root. Verify it exists.

Slow Performance / Long Videos

• Deepfake detection is compute-intensive. Running on CPU (especially in Docker) will be slower than native execution.
• Smart Sampling: For videos longer than ~50 seconds, Sentinel automatically uses "Block Sampling" to analyze distributed segments across the video duration, ensuring fast results without timeouts.vely on a GPU-enabled machine (e.g., Mac M-series with MPS).

---

Sentinel Initiative // Protecting Digital Integrity