Hacker News new | ask | show | jobs
by hackpert 1369 days ago
That is actually pretty freaking cool. Not hard to do by any means, just cool.
2 comments

The machine learning element isn't really needed if the instagram posts are tagged by date/time...
You could have taken the picture a day before or am I missing something? The post date/time itself does not mean much.
Many (most?) photos have EXIF tags with timestamps showing when the picture was actually taken, regardless of when it was posted. But given the amount of post-processing that likely happens to Instagram photos, I would not be surprised if that info was stripped out entirely, or at least altered.
Due to a massive rise in doxxing and stalking that occurred after 4chan made checking EXIF data an extremely common tactic when doxxing people - complete with "how to" guides for less tech savvy individuals - most image hosts and social media sites nowadays strip EXIF data from uploaded images. Some sites will detect/extract certain data only (eg: to include the camera/lens used for photography-focused image sharing sites) and scrub the rest. Been that way for roughly a decade if not longer now.
EXIF tags are scrubbed by most social media sites.
but they certainly keep the original with the tags somewhere.
Do you think the author is somehow gaining access to the originals? I do not see how this is relevant.
I think it's harder than it looks to do it reliably and automatically.
We have no idea as to the reliability of the system. We don't know how many false positives were found. We don't know if these were just hand selected from date/time/location.