search

davidsandberg / facenet

Face recognition using Tensorflow
MIT License
13.82k stars 4.81k forks source link

Question about cross validation #447

Closed JiaxiangChen closed 7 years ago

JiaxiangChen commented 7 years ago

I try to validate on my own data-sets. But when I look at the code I'm confused. In the facenet.calculate_roc, I see k_fold = KFold(n_splits=nrof_folds, shuffle=False) But KFold is used to split the dataset when using cross validation.I'm confused that I just want to validate my pre-trained model so I shouldn't split my dataset. Can someone help me understand this? Thanks advance.

davidsandberg commented 7 years ago

The distance threshold is selected using cross validation.

rustequal commented 4 years ago

The distance threshold is selected using cross validation.

I'm sorry, but I have the exact same question. Can you tell me why we should use K-Fold to split the test dataset? The best distance threshold we can calculate without splitting the test dataset. We still do not perform the training operation in this calculations.