This repository contains the implementation of our visual place recognition model for geolocalization, building on the work by Ali-Bey et al. The project focuses on two primary enhancements:
- Adjusted Miner Behavior: Reclassifying certain negative images as positive based on proximity to improve model robustness.
- Model Comparison: Evaluating various combinations of miners and loss functions to determine the best configuration.
- Code: In the Source folder can be found implementation of the visual place recognition model with both traditional and new methodologies.
- Results: In the Results folder can be found two Python Notebooks containing training and results of every model.
- Datasets: The datasets used for this project can be downloaded at this link.
Credits to the original implementation on which our work is based.