👁️ Eyedentify.AI – Conjunctivitis Detection Web App

This is a real-time web-based screening tool that detects signs of conjunctivitis (pink eye) using a webcam and deep learning.
It captures an image of your face, preprocesses it on the frontend, and uses a cloud-hosted model (via Hugging Face Spaces) to analyze each eye using Grad-CAM-enhanced deep learning.

---

🚀 Features

• 📸 Fullscreen camera interface with oval face guide
• 🧠 ResNet18 deep learning model with Grad-CAM++ visualizations
• 👁️ Eye detection powered by MediaPipe Face Mesh (via HF backend)
• ⚙️ Lightweight Flask frontend; inference offloaded to Hugging Face
• 🔐 All processing is user-controlled — no long-term image storage

---

🧰 Requirements

This app runs entirely in your browser + server (no training required)

Local Environment (Flask app only):

• Python 3.10 or 3.11 (❗ Avoid Python 3.12+)
• macOS, Linux, or Windows
• Pip

---

🛠️ Setup Instructions

1. Clone the repository

   git clone https://github.com/your-username/eyedentify-ai-app.git
   cd eyedentify-ai-app

2. (Optional) Create and activate virtual environment

   python3 -m venv venv
   source venv/bin/activate

3. Install minimal dependencies

   pip install -r requirements.txt

4. Run the Flask web server

   python app.py

5. Open in your browser:

   http://127.0.0.1:5000

---

🧪 How It Works

1. The browser captures your image and sends it to the Flask server.

2. Flask pre-processes the image and forwards it to the Gradio model endpoint on Hugging Face:

   • Detects facial landmarks
   • Crops and analyzes both eyes
   • Runs ResNet18 on each crop
   • Generates Grad-CAM heatmaps

3. The results — probabilities and visual explanations — are returned to the browser.

---

🌐 Cloud Model Hosting (Hugging Face Spaces)

The AI model is deployed at:

https://huggingface.co/spaces/luckyjain1/eyedentify-ai-model

This Space:

• Uses MediaPipe for facial landmark detection
• Crops and preprocesses eye regions
• Runs predictions with ResNet18
• Generates Grad-CAM++ overlays
• Returns all results via JSON

---

📁 Folder Structure

eyedentify-ai-app/
├── app.py                 # Flask server
├── templates/
│   └── index.html         # Main UI
├── static/
│   └── script.js          # Camera + interaction logic
├── requirements.txt       # Trimmed dependencies
└── README.md              # You're reading it

🧠 Technical Highlights

• Architecture: Decoupled frontend/backend — lightweight Flask handles UI, heavy inference runs serverlessly on Hugging Face.
• Explainability: Grad-CAM++ provides visual attribution maps for transparency.
• Efficiency: Eye crops are extracted from a single face using MediaPipe's landmark mesh.

---

📸 Sample Output

A Grad-CAM visualization will appear for each eye region, highlighting model attention on inflamed or discolored areas.

---