Urban-Rural-Classification

Streamlit App Link: click here

Introduction

Wanted to try out Streamlit. So I decided to implement a Rural-Urban Scene Classifier. Train and validation data downloaded from Kaggle. Manually downloaded test data from pexels.com.

How to use

• Refer to the demo in youtube.

• You can use data that I have in this folder by cloning this repo, or you can manually use data from internet.

• Test with data that are in similar grounds of training data.

• Upload only rural/urban scene images.

Implementation

• Picked up Inception-v3 Architecture for transfer learning.
• Finetuned it with urban-rural scene images by modifying the final classifier layer.
• Impressively, even after few epoch, able to get both train and validation accuracy to be 100%. Quite overfitted, will have to add more noise in augmentation step. But so far it predicts well.
• You can refer to my Jupyter file for implementation of transfer learning.

Improvements

• Incase certain images are misclassified please raise an issue along with the image.