GeoGuessr is a lot of fun. I played it a lot a few years ago, but only recently discovered that there’s a community of very serious players, and got interested in it again. For example check this video from yesterday where Tom Davies aka GeoWizard[1] gets a pretty good score on An Urban World: <a href="https://youtube.com/watch?v=TVt1GKBZMzc" rel="nofollow">https://youtube.com/watch?v=TVt1GKBZMzc</a><p>1: <a href="https://en.wikipedia.org/wiki/GeoWizard" rel="nofollow">https://en.wikipedia.org/wiki/GeoWizard</a>
I started working on a similar project - my approach was to train a model based on all of the geoguessr contest sets.<p>Geoguessr gives you both the street view _and_ the correct answer when you guess wrong.<p>They're all hand-curated so there are only so many points that you need to identify (the largest sets are 100k+ points) so the idea would be to run through it thousands of times with wrongs answers but use the correct answer to train the model<p>It's only a matter of time before the game has cheat bots and that problem, much in the same way online chess has that issue
I'm a GeoGuessr fan, and like the author, experienced a bit of a GeoGuessr renaissance when I had a baby. I guess it's the sort of slow, quiet game that lends itself to being played between baby care tasks.<p>In any case, the criteria discussed here is the sort of thing that all too often makes me skeptical of deep learning results.<p>There are the things that I consider "real" discriminators when playing the game - architectural styles, building materials, road quality, driving direction, alphabets/languages on signage, flags, flora, topography, makes/types of vehicles, skin color and clothing style of humans, etc. Then there are the "artificial" things - glimpses at the street view vehicle, copyright notices, image quality, knowledge of which countries do or don't have Google Street View coverage, etc.<p>Obviously by taking what I'm calling the "artificial" criteria into account a deep learning approach would score better today than if only the "real" criteria were considered. But I feel like if tomorrow, GeoGuessr swapped Apple Maps Look Around or Bing Maps StreetSide in place of Google Street View (or if Google released vastly updated imagery), the deep learning approach would fall apart, but humans who have built knowledge of the "real" features to look for would continue to do just about as well.
Interesting post, but I disagree with the conclusion that with a bit of work AI could take down the best Geoguessrs. The best players already take into account all the metadata like camera quality and distinctive features of the Google car. Furthermore, they would surely have an edge in urban locations where proximate locations are visible in street-signs. I could be proven wrong though, and it would be super interesting to reverse engineer AI guesses that turn out to be surprisingly correct based off little info.<p>To see just how good the top players are at instantly recognising a country using these clues, then check out <a href="https://www.youtube.com/watch?v=zEmoAYpTJuA" rel="nofollow">https://www.youtube.com/watch?v=zEmoAYpTJuA</a> . Or maybe don't if you don't want the game spoiled!
I wondered if this could be useful for open source journalism stuff like bellingcat does. Essentially combing through tons of photos online to figure out where / when they were taken and Put together an understanding of an incident
You know how you get off a plane and everything about a place 3000 miles away feels different? Not just smells and heat, but the light seems different. The surfaces and materials are different. The uh...(actual) atmosphere seems different.<p>Is for example, the atmosphere seeming different a real physical phenomenon that could be detected by ML?
Ooh, this is great. I love GeoGuessr and have pointed several people to it. I have thought that a trained AI could do better than me on average.<p>I've been posting my runs publicly to Strava for years. I recently started living in a small town and decided to stop posting my location, and using an open source tool called RunnerUp to track my runs. I have still been posting running photos that don't reveal my location. I know someday people will be able to figure out my location despite avoiding taking pictures with signs and stuff. If anyone wants to try to figure out one of my recent locations, my instagram link is in my bio.
This could be useful for finding new places to go that are look similar to your holiday snaps.
Also if I could do this with a picture of food I want to eat.. and it guess the probable location of a restaurant serving it, that would be great.