about the result about facenet

[davidsandberg]

I have tried to use facescrub with ~500 different people and ~80000 images in total. I'm not sure how well this model performed on LFW, but a link to the model can be found on the readme page. But perhaps you need to use an older version of facenet to be able to load the model. I don't think the model performed much better than 0.66, so my feeling is that your performance is alright.
There is nothing in the model that requires a certain ordering of the color channels, but of course the same ordering has to be used for both training and testing.
To improve performance i think the best way forward is to try with a larger dataset. For example it would be interesting to try the VGG Face dataset (http://www.robots.ox.ac.uk/~vgg/data/vgg_face/).

[sunjia2015]

Thank you for your quick reply.
I will try to use an bigger dataset to train the network and let you konw the result.