Twitter hates this Russian startup’s facial recognition software, apparently with good reason
With up to 10% of all accounts on Twitter being fake, that’s a lot of chaff to sort through to determine who is real and who is a troll or bot. FindFace, also available as an Android app from a contracted development studio, Trinity Digital, is heavily advertised for Twitter searches as a way to catch fake profiles, that is, accounts most likely using stock photos or altered versions of celebrities and politicians.
It also says it can help you locate missed connections on the platform with the facial recognition software developed for it by NTechLab, an up-and-coming Russian biometrics company.
However, this rollout hasn’t pleased Twitter one bit. Although capable of combing through this data, there is no official partnership with Twitter and these other companies. “When their use of our data was reported to us, we determined that it was a violation of our rules and we promptly took action,” a Twitter spokesperson told BuzzFeed.
The site, findface.me, initially asked me to share Twitter account information with it, but that option is no longer on the table due to the ban. NTechLab says it did not contact Twitter in advance, and only has access public profiles right now.
Questions for further clarification from Twitter were not returned at the time of this writing, including if it would take further action against the company, or what (if anything) NTechLab could do to ameliorate these concerns in line with Twitter’s new developer policies.
Those policies were introduced following revelations that US law enforcement and intelligence agencies were pulling data from Geofeedia to carry out surveillance. Twitter, as well as Facebook, cut off Geofeedia’s access in response to the public backlash. And this month, Twitter had Dataminr, an analytics service linked into its “firehouse” of data, end cooperation with those federal, state, and local offices over privacy concerns.
No masks here
NTechLab’s advanced FaceN software won big at the 2015 MegaFace Benchmark contest, and then appeared, to much consternation among users, on Vkontakte (VK), a Russian social media platform comparable to Facebook. The accuracy, and public quality of the site, meant that it was an inadvertent boon to stalkers, especially those targeting women. That, along with the sometimes cavalier attitude of its founders to such concerns, definitely soured Twitter on the whole affair.
Despite such concerns, Lenta.ru reported in May that FindFace had already exceeded a million users, and that the larger recongnition industry will be worth $6.2 billion by 2020. According to a PWC survey, “52% of adults agree that automated facial recognition will replace the need to remember names.” Beyond regular consumers, NTechLab has an especially bright future in law enforcement applications. It is hoping to expand its business in a very competitive market, as such systems are common in Russia through “safe cities” programs.
CCTV cameras in Moscow already employ NTechLab wares – in one instance, helping to ID arsonists – and casinos are also interested, to help catch cheaters and deadbeats. And it has been further tested at public events like concerts, able to locate attendees quickly. NTechLab also expects a lot from a “strategic agreement” with TIAR Technology, which works with the Turkish biometric security vendor Papilon Savunma, whose products are especially popular for prison record keeping and are sold worldwide.
NTechLab co-founder Alexander Kabakov said in a press release that “FindFace will help Twitter users ensure their photos aren’t being used for fake accounts. On top of that, users can find people they knew previously, from old relatives to old relationships. When an exact match isn’t found, FindFace will identify the most similar matches instead.”
Smile for the camera
Wanting to know for ourselves if FindFace’s claims were up to scratch, Geektime tested the service to see exactly what it could tell you about Twitter users.
Users submit a photo to the site, which maps out facial geometry and creates a “face print” to search for images on the site. FindFace says it stores some of the data for a limited period of time, but that it does not “sell or rent” the prints to third parties, and that the photos will auto-delete once the search results are compiled.
Users must also be 18 years old to use it, and the app scans 313 million profiles to return the “top similar 1000 Twitter profiles’ pictures” that come closest to matching the upload.
Testing the device on two famous people (Barack Obama and Britney Spears), myself, and a randomly-selected stock photo of a man, I found that it generally does what it sets out to do with finding potentially fake profiles. It at the very least is able to detect when stock photos have been used, even if altered, and can in the case of the president and Britney, find variant images of the person.
As my picture was an older one back when I had a beard, though, it failed to find any exact match. The reference points it did return were just generally similar: Big forehead, thin round glasses, full facial hair. Still, if you were looking for someone like me with a photo from, say, a party a few years back and trying to put a name to that face, you’d have a pretty good start with these results.
(I never post personal photos on Twitter, so really, it’d require Facebook to really deliver in my case. But then, that’d be true for a lot of people and is why the service has been so unsettlingly accurate on VK.)
It was able to find non-photographic results as well, though with mixed results. The Mona Lisa, for instance, returned hits, but a search of cartoon characters from 1990s TV shows didn’t work so well as they were not obviously human enough to be read well.
Despite Twitter’s hostile reaction, the company continues to hope for a formal relationship. For now, you can still get results, despite Twitter’s ban on FindFace. NTechLab also says it is in talks with potential US partners, but declined to name them.