Landmark recognition using deep learning

Abstract

The project assigned was Visual Recognition using Deep Learning. Specifically, this project aims to explore the possibility of using deep learning techniques to construct an image classifier and a text-based image retrieval system.

The first part of this project will focus on the construction of a landmark classifier. Firstly, landmark images will be gathered manually from the internet to construct the required datasets. Transfer learning techniques will then be utilized to train a landmark classifier. Several different models and hyperparameter settings will be studied and tested. The best model with the best settings will be selected as the final classifier. The classifier will then be used to classify and label images that have landmarks in them.

The second part of this project is then to explore the possible practical applications of the trained landmark classifier. Specifically, the trained classifier will be used to construct a text-based image retrieval system to allow the classified and labelled images to be retrieved via text search. Simple GUIs will also be designed and constructed to produce a user-friendly final product.

The performance of the classifier and retrieval system will then be tested using various performance metrics.