How can i get model?

[neuralix]

Thank you for offering good code.

I have done training, and get files like below.

in {OpenfaceRoot}/training/work/21: classes.t7, model#.t7, optimState_#.t7, opts.t7, test.log, train.log

How can i get corresponding files(:{celeb-classifier.nn4.v1.pkl, nn4.v1.t7, shape_predictor_68_face_landmarks.dat}) from files above?

[sschnug]

Your "training" is only responsible for training the model (the neural network). So your training replaces "nn4.v1.t7" (as i understand it).

The other files are not generated during training:

• shape_predictor_68_face_landmarks.dat is trained data for landmark-detection in faces commonly used in dlib (training is not part of this project)
• celeb-classifier.nn4.v1.pkl seems to be a pickled file describing a trained classifier on some images (whose features were extracted with the nn4.v1 network) -> you can generate your own classifier (read documentation). A training of a classifier processes the feature-vectors of the images, therefore you have to use your trained network to generate these and then build your classifier!

Remarks:

• Obtaining these files is explained here link to docs
• You should already have these because i think you will need the landmark-detection model for preprocessing your images prior to your training!

[bamos]

Thanks @sschnug for the great references!

@neuralix - the trained DNN model indeed replaces nn4.v1.t7 if that's what you want to do. Make sure you understand the differences between training a DNN model and a classification model, which the documentation describes:

• http://cmusatyalab.github.io/openface/demo-3-classifier/
• http://cmusatyalab.github.io/openface/training-new-models/

I'm closing this since the issue tracker is for development discussions and tracking bugs, but please come to the Gitter chat or mailing list (linked in the README) if this wasn't clear.