Skip to content

Latest commit

 

History

History
16 lines (11 loc) · 1.1 KB

File metadata and controls

16 lines (11 loc) · 1.1 KB

StreetNet

AI trained to play country streaks on GeoGuessr. When placed in a random location on Google Street View, this neural network is tasked with predicting which country you are in.

You can test StreetNet by playing a game of GeoGuessr, or by simply dropping into a random location on Google Maps!

Screenshot of the StreetNet GUI running while in a game of GeoGuessr, where the location was correctly identified as Finland.

StreetNetInFinland

How To Use

To run, simply clone this repo, install the requirements (requirements.txt or cuda_requirements.txt) and run with python:

python streetnet.py [--use-cuda] model_path

Currently there are 2 trained models included in this repo: model_18.pt and model_50.pt. These use the ResNet-18 and the ResNet-50 architectures respectively. model_50.pt achieves a higher accuracy, however it can require up to 10.6GB of memory so you may prefer using more light-weight version - model_18.pt.