Facial Landmarks Meaning

edit

When creating a facial landmark dataset, a researcher has to go through many facial images and use tools to annotate the locations of facial landmarks in the images. How many landmarks and the order of the landmarks the researcher annotates must be same for each image. That way, a model can be trained using the annotated images in order to detect the landmarks given a new image.

Therefore, the facial landmarks that the points correspond to (and the amount of facial landmarks) that a model detects depends on the dataset that the model was trained with.

The pretrained FacemarkAAM model was trained using the LFPW dataset and the pretrained FacemarkLBF model was trained using the HELEN dataset. Each of these datasets use the same 68 landmark configuration shown in the image below: