model definition selection

davidsandberg / facenet

Face recognition using Tensorflow

MIT License

13.75k stars 4.81k forks source link

Open zhenglaizhang opened 6 years ago

zhenglaizhang commented 6 years ago

Hi,

Currently inception_resnet_v1 is used to train facenet, I notice definition for inception_resnet_v2, and nn2, nn3, nn4 are also provided.

Is currently inception_resnet_v1 providing best performance even than NN2 network?

thanks in advance!

davidsandberg commented 6 years ago

The resnet type architectures are easier to train. Except from that it's mainly a question of training speed.