Knowledge Distillation for Satellite Image Classification

This project demonstrates knowledge distillation applied to the EuroSAT RGB dataset using TensorFlow and a VGG16-based teacher model. The goal is to transfer knowledge from a large, accurate model (teacher) to a smaller model (student) for efficient inference.

🔍 Project Overview

• Dataset: EuroSAT RGB (via TensorFlow Datasets)
• Teacher Model: Pretrained VGG16 (frozen convolutional base)
• Student Model: Custom lightweight CNN
• Loss: Combination of categorical crossentropy (hard labels) and MSE (soft labels)
• Goal: Improve efficiency and accuracy.

🚀 How to Run

1. Clone the repo:

git clone https://github.com/hardikmakkar07/knowledge-distillation.git
cd knowledge-distillation