search

ageitgey / face_recognition

The world's simplest facial recognition api for Python and the command line
MIT License
52.91k stars 13.44k forks source link

How did you train this model? #481

Open hoangdzung opened 6 years ago

hoangdzung commented 6 years ago

Your model works very well. I really wanna know which technique you used to train this embedding? Did you use triplet loss or any other special technique? Thank you.

cftang0827 commented 6 years ago

The model of face_encodings is from dlib's metric learning sample code

http://blog.dlib.net/2017/02/high-quality-face-recognition-with-deep.html

so if you are interested in it, please have a look of dlib's code http://dlib.net/dnn_metric_learning_on_images_ex.cpp.html

Thanks

hoangdzung commented 6 years ago

Which dataset did you use for training? Public dataset like VGGFace2 or dataset of your own? Your model performs quite well, better than facenet model. But the performance hasn't meet my requirement, so I wanna retrain this model with people faces of my country using your model as a pretrained model.

ageitgey commented 6 years ago

There's a description of the dataset here: https://github.com/davisking/dlib-models#dlib-models