3 ms·
As Wired writes, it is a conceptual piece. I see the vision technology comes from Polar Rose. I've worked with them for a couple of years as their lead architec
by st3fan 17y ago
As Wired writes, it is a conceptual piece. I see the vision technology comes from Polar Rose. I've worked with them for a couple of years as their lead architect. Until they funding dried up. I think they are running on fumes now.
Vision technology is really tough to sell. Finding faces is simple, everybody is doing that these days, but actually recognizing a face is a whole different game. It is really hard and you need a lot of cpu power and storage to match someone in near realtime. At Polar Rose we got good results with stock photos. But throwing 'internet quality' photos at it resulted in just terrible accuracy.
What they do in the demo cannot work just on the phone. It will need a support infrastructure that receives the mugshot, processes it and then matches it against other faces in the database. Nobody really gets this part right yet.
- wizard_2 17y agoI used to develop some facebook apps. I think in this aspect you could process your network (and maybe their networks?) pull all the photos of those people and look for their face. Facebook doesn't share the "tagging" data afaik, but you could build a limited library of who you're going to be taking photos of. So even if the recognizing isn't very accurate it wouldn't have to be precise.
- abyssknight 17y agoI was once approached to help build something like that for a prison. Essentially, it'd match the prisoner to their mugshot to ensure the right person was let out. I turned the position down, as when I asked about how they would store the eigen-vectors for that many people and index them I was met with an awkward silence. I do like the idea of processing Facebook's data, but that's another plan entirely.