locus-model

LOCUS aims to automate GeoGuessr with Deep Learning. This repository contains the code for the Deep Learning model.

Setup

Setup environment

First create a virtual environment and activate it:

make create_environment

Then install the requirements:

# to install requirements
make requirements

# to install requirements and dev requirements
make dev_requirements

The project currently uses the pytorch nightly build for Python 3.12 compatibility. We will move over to the stable build upon full Python 3.12 support.

Setup data

The dataset is found on (kaggle)[https://www.kaggle.com/datasets/habedi/large-dataset-of-geotagged-images]. Extract the dataset to the data/raw/LDoGI folder.

To preprocess the data, run:

make data

Setup database

install docker

install postgres docker

create network

    docker network create db

create postgres container

docker run -d --name pgadmin --network=db -p 80:80 -e PGADMIN_DEFAULT_EMAIL={YOUR_USER} -e PGADMIN_DEFAULT_PASSWORD={YOUR_PASSWORD} dpage/pgadmin4

run psql

docker run -it --rm --network=db postgres psql -h locus-db -U postgres

run database

docker run --name locus-db -p 5432:5432 --network=db -v "$PWD:/var/lib/postgresql/data" -e POSTGRES_PASSWORD={YOUR_PASSWORD} -d postgres

Project structure

The project structure can be seen in the structure.md file.