Photo-Geolocation-Recognition Based on Covolutional Neural Network
Our project focuses on geolocation identification through detecting architectural styles using Convolution Neural Networks (CNN). In our project, we study both landmark and non-landmark buildings, as both architectures encode lots of valuable information about places. Non-landmark buildings of each city are also a major part of its culture and defines it. Therefore, learning architectural styles at a larger scale can contribute notably to solving geo-localization problems on general street-view photos.
The goal of our project is to better utilize the architectural-related content and extract architectural features which can benefit the general geolocation recognition process.
We created a dataset consisting of building images of nine cities from Flickr.com. Then, we implemented different convolutional neural networks for recognizing architectural styles of landmark and non-landmark buildings, and compared their performance on identifying cities with architectural information.