search

ageitgey / face_recognition

The world's simplest facial recognition api for Python and the command line
MIT License
52.88k stars 13.44k forks source link

Relationship with OpenFace #514

Closed chuyee closed 6 years ago

chuyee commented 6 years ago

Hi Adam,

Thanks for your great work! I have a quick question on the relationship with the CMU OpenFace. According to your Medium post 2 years ago [1], the original face_recognition is designed to simplify the OpenFace interaction.

UPDATE 4/9/2017: You can still follow the steps below to use OpenFace. However, I’ve released a new Python-based face recognition library called face_recognition that is much easier to install and use. So I’d recommend trying out face_recognition first instead of continuing below!

Is it still true for now? AFAICS, the current face_recognition NN is purely relied on dlib (Is it correctly?) If so, what's the motivation for the switch? Better accuracy? Less recognition time? Easy of use? Thank you and hope you can answer my question.

Thanks, -yi

  1. https://medium.com/@ageitgey/machine-learning-is-fun-part-4-modern-face-recognition-with-deep-learning-c3cffc121d78
ageitgey commented 6 years ago

Yep, this library purely relies on dlib. The motivation for the switch was ease of use and ease of installation. Of course OpenFace still exists and you can use that instead if you like.

chuyee commented 6 years ago

Thanks Adam. I did some search in the web and seems like dlib dnn does have a better LFW score than OpenFace now. [1][2].

  1. http://blog.dlib.net/2017/02/high-quality-face-recognition-with-deep.html
  2. https://cmusatyalab.github.io/openface/models-and-accuracies/
ageitgey commented 6 years ago

LFW is a little bit of a synthetic benchmark. It's weighted towards western celebrities. You can always try out both and see what works better for your specific use case :)