Undocumented Facebook API to identify friends in photos(narenonit.blogspot.com) |
Undocumented Facebook API to identify friends in photos(narenonit.blogspot.com) |
It's a giant pain to screenscrape this using 'curl'. If I recall correctly, the bounding box coordinates I wanted are set as CSS properties inside inline HTML sent to the client wrapped up in a Javascript string literal as part of Javascript served to the client as the result of an AJAX call, if memory serves correctly. To get my screenscraper working, I had to do the AJAX call, parse the literal javascript, walk the AST to find the string literal I needed, parse the HTML to find the element I needed, then use the computed CSS properties. Looks like the author of this post found a much nicer way.
(note: that work wasn't about recognition; it was about just finding the faces in images, not identifying them)
If I hit yes, I'm tagging friends who might not want to be tagged. Furthermore, I might end up in the same boat with friends tagging pictures of me! Either way, I help better Facebook's facial recognition, which unnerves me.
On the surface, clicking "no" means that they got the facial recognition wrong. But what am else am I revealing? If the match was 98%, would they infer that one of us (or both) is concerned about privacy? That we have something to hide?
The third alternative is to click nothing. The only information that gives Facebook is that I'm not interested in helping curate their data any more than I already am.
People gravitate towards Selenium too quickly, when you really only need Selenium to test rendering.
> I found this API has a limitation it only suggests people in the photo who are in your FB friend list
Face.com did facial recognition on Facebook with no limits back in 2010.[2] Then Facebook bought them and killed the broad face recognition.[3]
There are other face recognition companies now, but they're keeping a lower profile about how broad their database is. Except for Findface[4] in Russia. They loaded up the entire photo database of Vcontact, a social network in Russia. 70% success in identifying random young people on the St. Petersburg metro.
[1] https://www.epic.org/privacy/facebook/Facebook-Biometric-Rul... [2] https://web.archive.org/web/20100701083622/http://face.com/ [3] https://web.archive.org/web/20120723211743/http://face.com/ [4] http://findface.ru/
This http://arxiv.org/pdf/1412.1897v4.pdf could be one way. You could generate images which are not of someone who wants to protect his privacy, but tweak them to strongly correlate to an image of that person by the neural net, whereas it could be a picture of anything you have just optimized (the images could be just noise, or another person, etc). You could generate batches of those and upload them, and confirm to Facebook that they are indeed photos of the person who wants their privacy back. You could repeat this (automatically) and corrupt the weights in FB's neural net, which would overcome their face detection abilities for the individual in question.
If I were Facebook, I'd look at the entropy of an image before attempting to classify it. Anything particularly high (noise) or low (squiggles) would be discarded before running through the classifier.
I'm still just as vulnerable to attack, if not more (ie: trying to host my own email service). The only difference is I'm markedly more isolated from my friends. It simply wasn't worth the trade off.
It's similar to people who get ridiculed for stockpiling guns and food in underground bunkers for the coming War-for-Independence 2.0. Sure, I could become a "digital prepper" and survive the data-pocalypse--likely at the cost of my relationships.
It's not like a "10 Cloverfield Lane" lifestyle is particularly appealing to me, either. I'll live with my friends for now, with only a mild sense of paranoia.
I say this as someone who has exercised the option to avoid Facebook entirely for the past 7 years (including noscript+faceblockers for the pervasive thumbs-up buttons.
Just make sure you look up the DOI on libgen.
The only difference that I imagine exists is being able to search from the pool of "everyone" rather than "your friends", but that would make it less confident. In the law enforcement version, there would probably be a way to anchor the search to another individual just to improve accuracy to an acceptable level.
To be concrete, my local supermarket and Amazon have my full name, and partial purchase information (the supermarket via a loyalty card). Both have ways to contact me with promotions. You don't think they'd like to know what I like on Facebook (if I did that) or what I search on Google that might suggest purchase intent?
I think they'd love that data, the platforms just make more money by only allowing them to target with it.