search

TadasBaltrusaitis / OpenFace

OpenFace – a state-of-the art tool intended for facial landmark detection, head pose estimation, facial action unit recognition, and eye-gaze estimation.
Other
6.85k stars 1.84k forks source link

Any possibility of real-time multiple face recognition? #908

Open ECas95 opened 4 years ago

ECas95 commented 4 years ago

Hi! I'm making a paper about security in stadiums using facial recognition, and I'm thinking on using OpenFace for it, but my question is if the solution have multiple face recognition at real time support,

xkortex commented 4 years ago

I'll let the author have the final say in case he knows about something clever I don't, but as someone who has worked extensively with facial recognition pipelines, there's nothing off-the-shelf that you can just spin up and run and get real time 30 fps facial recognition.

You can get decent results at 1-5 FPS depending on the framework.

This library also doesn't currently have any kind of API integration so I don't think it matches your use case.

lavrenkov-sketch commented 3 years ago

@xkortex

You can get decent results at 1-5 FPS depending on the framework.

hi, can you tell me which frameworks will help achieve such a result if eating a large number of 1000+ classes

TadasBaltrusaitis commented 3 years ago

There are typically several steps to face recognition:

  1. Detecting the faces in the image (and possibly landmarks), this is what OpenFace can be used for or help with
  2. Doing the actual face recognition (or verification, depending on the actual use case)

Both of them have a computational cost associated with them. And I agree that it would be tricky to get 30fps, but a question would be, why do you need 30fps for face recognition, unless people are moving super fast in the scene, you should be ok with lower fps and some additional tracking.