Introduction

This is a simple class project that uses VLAD feature to build a large scale VPR (Visual Place Recognition) system. Given a query, you can retrieve images similar to the query from a database of images.

Implementation

More details can be found in the markdown section of the notebook.

Requirement and Run

Besides some of the common default Python package,you will need numpy, scikit-learn and OpenCV >= 4.4 as the code uses SIFT feature detector that's only available after that specific version.

To run the code on any database of images and with customized query, replace the database_path and query_path to different folders and run the entire notebook. It may takes a while to index all the images in the database.

Currently the code will return the top 1 similar image of a given query. To increase this number, change num_of_imgs accordingly.