-
Notifications
You must be signed in to change notification settings - Fork 1.9k
Model Training
There are a number of trained models that OpenFace uses and the code for training all of them is available.
OpenFace uses the Constrained Local Neural Field model for facial landmark detection, it requires three trained modules - Point Distribution Model, Patch Experts, and landmark verification module.
To train the statistical model of shape go to:
matlab_version/pdm_generation and read the readme file there
More details here in Chapter 4.2 specifically 4.2.5
You can find the patch training code here - https://github.com/TadasBaltrusaitis/CCNF
More details about the method itself here in Chapters 4.3, 6.2, and 6.3 and here
A module for validating face detections (training and inference), it is used for tracking in videos so as to know when reinitialisation is needed:
matlab_version/face_validation
Follow the readme there for instructions
More information about Facial Action Units can be found here - https://github.com/TadasBaltrusaitis/OpenFace/wiki/Action-Units
To train Facial Action Unit detectors follow the instructions in the readme file:
matlab_version/AU_training
More details about the method used here