🚀 AuthenSight AI

Deepfake detection system for images and videos, built during a hackathon at IIIT Una.

AuthenSight AI analyzes facial regions using a ResNet-based deep learning model and exposes prediction APIs through a scalable REST architecture.

---

🧠 Overview

AuthenSight AI detects AI-generated or manipulated faces in:

• 🖼 Images
• 🎥 Videos (frame-level inference)

The system extracts faces, preprocesses them, and runs inference using a pretrained deepfake detection model built on ResNet.

---

🏗 Architecture

Frontend (React)
        ↓
Express.js (API Layer)
        ↓
FastAPI (ML Inference Service)
        ↓
PyTorch (ResNet Model)

---

🔧 Tech Stack

Machine Learning

• PyTorch
• ResNet-based classifier
• OpenCV (face extraction)

Backend

• FastAPI (AI inference server)
• Express.js (API gateway)
• RESTful architecture

Frontend

• React.js

---

⚙️ Features

• Image deepfake detection
• Video frame sampling & aggregation
• REST-based API endpoints
• Real-time inference
• Modular backend design
• Scalable architecture

---

📦 Installation

1️⃣ Clone Repository

git clone https://github.com/shauryaa108/authensight-ai.git
cd authensight-ai

2️⃣ Create Virtual Environment

python -m venv venv
source venv/bin/activate

3️⃣ Install Dependencies

pip install -r requirements.txt

---

▶ Running the FastAPI Server

uvicorn model.app:app --reload

Swagger Docs:

http://127.0.0.1:8000/docs

---

📡 API Endpoints

🔹 Image Prediction

POST /predict

Form Data:

queryObject: <image file>

Response:

{
  "label": "FAKE",
  "confidence": 87.23
}

---

🔹 Video Prediction

POST /predict-video

Form Data:

queryObject: <video file>

Response:

{
  "label": "REAL",
  "confidence": 91.12,
  "frames_analyzed": 34
}

---

🧪 How It Works

1. Extract largest face from image/video frame
2. Preprocess face
3. Run ResNet model inference
4. Compute fake probability
5. Aggregate results (for videos)

---

⚠ Limitations

• Frame-level detection (no temporal modeling)
• Works best with clear frontal faces
• Accuracy depends on face extraction quality
• Not a forensic-grade system

---

👥 Team

Built during a hackathon at IIIT Una.

• Rishik — ML & FastAPI
• Yash & Aman — Frontend (React) & Backend (Express.js)
• Yash & Aman — Frontend (React) & Backend (Express.js)

---

📌 Future Improvements

• Temporal deepfake detection models
• RetinaFace integration
• GPU batching
• Confidence calibration
• Cloud deployment