Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

Just got doxxed to within 15 miles by a vision model, from only a single photo

twitter.com

Post reply on HN
Print view View original

41–50 of 67 posts

Re: Just got doxxed to within 15 miles by a vision model, from only a single photo

#41

Post by drbig »

mastermedo wrote: post #21

Original link https://geospy.ai/ . Pretty scary stuff, I took a picture of one tree, got pinpointed to 4 km.

Um, no. Did a test and the results are neither scary nor wowing me now:

1) Easy mode: Cityscape of Naples, Italy; Bot 100% match, nice, I guess.

2) Mid-mode: A road in Glenoak Hills, CA; Bot says Malibu, CA... Never been there.

3) Hard-mode: Main lake in Ostróda, Poland; Bot says it's in Manhattan and that the vegetation matches Northern American :-D Ahahaha, didn't even get the continent right so...

Note: Tested on personal photos, with EXIF stripped, not widely/at all uploaded to the nets.

So... perhaps works great if you feed it back stock photos you've grabbed from the first result of a google search? Because my mileage is barely an inch.

Re: Just got doxxed to within 15 miles by a vision model, from only a single photo

#42

Post by thadk »

Beyond the general look of places or particular constructed tells, I wonder if the model is taking any hints from analemma type patterns or latitude by the sun position (e.g. https://tvtropes.org/pmwiki/pmwiki.php/UsefulNotes/TheLongit... ). Cryptonomicon's use of that problem came to mind.

Re: Just got doxxed to within 15 miles by a vision model, from only a single photo

#43

Post by qrios »

smartbit wrote: post #36
mastermedo wrote: post #21

Original link https://geospy.ai/ . Pretty scary stuff, I took a picture of one tree, got pinpointed to 4 km.

Took a picture and it recognized the architecture style, guessed the wrong town 60km away.

Same here: Picture from rural Austria in the North and it comes up with Hungaria. Location is in the center of Budapest, 300km distance to the original view point. But architectual style looks really hungarian.

Re: Just got doxxed to within 15 miles by a vision model, from only a single photo

#45

Post by sshine »

RGamma wrote: post #39
mastermedo wrote: post #21

Original link https://geospy.ai/ . Pretty scary stuff, I took a picture of one tree, got pinpointed to 4 km.

Gotta wonder why you'd even develop that. This thing seems to have essentially no value for for anyone but bad actors?

Imagine: you’re abducted, but manage to sneak a photo and upload it somewhere completely random with a message for help. You don’t even know where you are.

Re: Just got doxxed to within 15 miles by a vision model, from only a single photo

#46

Post by tgv »

quietfox wrote: post #23
tgv wrote: post #17

So, someone, probably a group, made the effort to build a system that can infer location from a picture. The negatives are obvious. What possible beneficial uses can it have?

Investigative journalism like the OSINT based researches conducted by bellingcat would be one example, as long as the AI results get doublechecked for accuracy.

They could get geoguessers to participate, I guess. No need to build a publically accessible model.

Re: Just got doxxed to within 15 miles by a vision model, from only a single photo

#47

Post by eqvinox »

maxglute wrote: post #30
eqvinox wrote: post #24

Earlier quoted context omitted.

Satellite view? This is about street-level photos… I don't think you would mix these two things in the same model, it'd just dilute the capabilities on both.

It identified aerial shots of hawaiian islands, prince edward island, cuba and most of the commonly "known" islands with distinct shapes fine. At least much more consistently than random street level views I've thrown at it, where description seems sensible, in as much as a good geoguesser can descripe their rationales.

I don't question that some models can also handle satellite views, just that it makes little sense to "pollute" a model intended for geolocating photos with input data from a different category. Especially since satellite views are just vaguely similar enough to photos that I'd assume they can cause quite a bit of unnecessary ambiguity and model complexity to handle it.

Re: Just got doxxed to within 15 miles by a vision model, from only a single photo

#48

Post by BrandoElFollito »

Two photos (of rather distinctive places, same street), each one localized hundreds of kilometers away. At least the country matches.

I guess that this is really hit and miss depending on a bunch of parameters.

I wonder how Street View would perform on signatures such as Shazam uses

Re: Just got doxxed to within 15 miles by a vision model, from only a single photo

#49

Post by bojan »

RGamma wrote: post #39
mastermedo wrote: post #21

Original link https://geospy.ai/ . Pretty scary stuff, I took a picture of one tree, got pinpointed to 4 km.

Gotta wonder why you'd even develop that. This thing seems to have essentially no value for for anyone but bad actors?

It can do wonders for OSINT and disproving propaganda (for what's that worth).

Re: Just got doxxed to within 15 miles by a vision model, from only a single photo

#50

Post by RobinL »

RGamma wrote: post #39
mastermedo wrote: post #21

Original link https://geospy.ai/ . Pretty scary stuff, I took a picture of one tree, got pinpointed to 4 km.

Gotta wonder why you'd even develop that. This thing seems to have essentially no value for for anyone but bad actors?

Feels like there could be anti fake news applications. It's pretty common for photos to be recycled and purported to be of current events, when what they actually show is far away both in distance and time
Post reply on HN