Lost & Found: Predicting Locations from Images

General

This is a project for HSLU (DSPRO 2).

It consists of a codebase, a scientific report (source), and a published dataset.

The project uses Markdown (and LaTeX), Python and Node.js with Typescript.

It uses Obsidian (Markdown) for documentation and planning, it can also be viewed with any other markdown viewer (including GitHub/GitLab).

For Python dependency management it uses poetry.

To set it up do poetry install, to add dependencies use poetry add.

To run commands use poetry shell to spawn a subshell.

Select the venv after running poetry install for Jupyter Notebooks.

For Node.js dependency management it uses yarn v1.

To set it up simply type yarn.

To see available commands, check out the scripts section of the package.json and run them using yarn <command>.

All project relevant commands are handled via yarn, including formatting our Python, Typescript, and Markdown files and generating our report from our Markdown source.

Loading data from our published dataset

To load the data from our published dataset:

• Ignore our server check (DOWNLOAD_LINK=None and SKIP_REMOTE=true in .env)
• Put the unzipped data(_mapped) directory (if you want both start with mapped) into dspro2/1_data_collection/.data.
• Run yarn data:import on a unix based system (or rename them to geoguessr_location_******.png and geoguessr_result_******.json, and copy all the JSON files into dspro2/3_data_preparation/01_enriching/.data).
• Execute the dspro2/3_data_preparation/99_importing/import.ipynb notebook (make sure to set the MAPPED parameter correctly and only the relevant data is inside the directory).

Collecting data

Simply run yarn scrape:prepare, set GEOGUESSR_EMAIL and GEOGUESSR_PASSWORD in your .env file, then yarn scrape:ui (for local testing), yarn scrape or scrape:deploy (for multiple parallel instances).

Course Coaches

Within this module we are supervised by the following course coaches:

• Dr. Umberto Michelucci
• Dr. Ludovic Amruthalingam
• Dr. Daniela Wolff
• Aygul Zagidullina (internal link)

Authors and acknowledgment

The whole project was done by the following students:

• Linus Schlumberger
• Lukas Stöckli
• Yutaro Sigrist