Questions in the procedure of validation on lfw

[zzw1123]

Thanks for your contribution. But I have a doubt that before feeding the testset into the model we trained already, should we first detect faces in the images of lfw?In order to make the embeddings contain only the part of face which is more pure?

[qixianbiao]

You should first detect and align faces for both CASIA and LFW. You can train the model and evaluate the performance on LFW each 1000 steps like Facenet does https://github.com/davidsandberg/facenet

I have provided three running logs of my method training on cleaned CASIA and evaluating on LFW below https://github.com/qixianbiao/PolyFace

[zzw1123]

Thank you,xianbiao. I remember that!