🌍 Lightweight Disaster Image Classification using Knowledge Distillation

This project presents a lightweight yet effective deep learning pipeline for disaster image classification. Leveraging knowledge distillation, we train a compact student model to mimic a larger, more powerful teacher model—making the solution efficient for real-world deployments on low-resource devices.

---

🔗 Live Demo

👉 Try the Streamlit App

---

🚀 Features

• ✅ Real-time image classification via web interface
• 🧠 Knowledge distillation for model compression
• 🔬 Supports multiple disaster categories (e.g., earthquake, flood, wildfire, etc.)
• 📦 Lightweight architecture optimized for deployment
• 🌐 Easy-to-use and accessible from any device with a browser

---

🧑‍💻 Tech Stack

• Frontend & Deployment: Streamlit
• Model Training: PyTorch
• Model Optimization: Knowledge Distillation
• Backend: Python
• Hosting: Streamlit Cloud

---

🗂️ Project Structure

.
├── app.py                   # Streamlit application
├── models/                  # Trained teacher and student models
├── utils/                   # Utility functions for preprocessing, prediction, etc.
├── assets/                  # Sample images and assets
├── requirements.txt         # Python dependencies
└── README.md                # Project documentation

---

🛠️ Setup Instructions

1. Clone the repository

   git clone https://github.com/zapfruit/Lightweight-Disaster-Image-Classification-using-Knowledge-Distillation.git
   cd Lightweight-Disaster-Image-Classification-using-Knowledge-Distillation
   

2. Install dependencies

   pip install -r requirements.txt
   

3. Run locally

   streamlit run app.py