serengil / deepface

A Lightweight Face Recognition and Facial Attribute Analysis (Age, Gender, Emotion and Race) Library for Python
https://www.youtube.com/watch?v=WnUVYQP4h44&list=PLsS_1RYmYQQFdWqxQggXHynP1rqaYXv_E&index=1
MIT License
14.34k stars 2.2k forks source link

Thresholds for FMR=0.1% #401

Closed sbnsl closed 2 years ago

sbnsl commented 2 years ago

Hi, Do you now about the threshold values corresponding to FMR=0.1% for FaceNet and ArcFace models using Euclidean L2 distance?

Thanks,

serengil commented 2 years ago

What FMR is?

sbnsl commented 2 years ago

I mean the threshold corresponding to false positive rate of 10^-3 on your validation set (e.g., LFW).

serengil commented 2 years ago

Actually, deepface wraps ready-to-use models. I mean that I haven't trained none of them for facial recognition. Those scores can be found in the reference studies or repos cited in the license section of the read me.