Accuracy on LFW, CFP and AgeDB30

[wjgaas]

this work looks great! i test it on the LFW, CFP and AgeDB30 with the model you provided, but got a low accuracy. i use the lfw.bin, cfp.bin and angedb30.bin in faces_emore, and test as follow: data = (data - 127.5) * 0.0078125 embeddings = model.forward()['View_1'] embeddings = sklearn.preprocessing.normalize(embeddings) Then, i got the results: [lfw]Accuracy: 0.99750+-0.00318 [cfp_fp]Accuracy: 0.95586+-0.00671 [agedb_30]Accuracy: 0.96983+-0.00864 with image flip: [lfw]Accuracy-Flip: 0.99733+-0.00281 [cfp_fp]Accuracy-Flip: 0.96057+-0.00783 [agedb_30]Accuracy-Flip: 0.97167+-0.00910

Is there anything wrong? or maybe you use images with more accurate aligned?