Image Retrieval for Visual Geolocalization: Extensions and Experiments

Official repository for the Machine learning and Deep learning project "Image Retrieval for Visual Geolocalization: Extensions and Experiments".

• The project is also available on [Kaggle] with all the datasets loaded.
• Training, Validation and Test can be run from main.py.
• Relative report is here.

Repository Structure

Datasets

The datasets folder contains the following scripts:

• Train.py: Class definition for the Train.
• Test.py: Class definition for the Validation and Test.

Models

The models folder includes the model architecture for the VPR pipeline.

• Aggregators.py: Script that contains the aggregation layers.
• Backbone.py: Script that contains a modified backbone of ResNet18.
• Evaluation_loop.py: Script that defines the Recall@K and the Evaluation loop.
• Training_loop.py: Script that defines the Training loop.

Visualizations

The visualization folder contains the file Visualization.py that defines functions to plot images and predictions.

Authors

The authors of the project are:

• Francesco Gaza s315489@studenti.polito.it
• Giovanni Monco s315001@studenti.polito.it
• Erika Spada s318375@studenti.polito.it